Table of contents of "The Protective Power": 21 chapters in 3 parts + prologue and 6 annexes | Adrian Potentela

This book was born of a concern that grew quietly, almost without my noticing. For years I have watched public conversation fragment, watched information become distorted, and watched citizens gradually lose their ability to find their bearings in an environment saturated with stimuli, narratives and manipulations.

It was not a sudden discovery but an accumulation of signals, patterns and small alarms that, seen together, reveal a structural problem: the cognitive sphere of our democracies is being eroded, and with it, the future of Europe as a global power.

I am not an academic or a specialist in constitutional law. My training does not come from universities or research centres. But I do possess a way of thinking that has marked my life: a systemic intelligence, an ability to see problems as interconnected systems, to identify their root causes and to understand that real solutions are not found in the symptoms but in the architecture that produces them.

This way of seeing the world has allowed me to detect something that, in my judgment, has not yet been addressed with the seriousness it deserves: the cognitive vulnerability of modern democracies is not just a risk; it is an existential threat to European leadership.

We live in an environment where information is no longer a neutral resource but the new geopolitical battlefield. Information manipulation, artificial polarisation, digital social engineering and the exploitation of cognitive biases have created a scenario in which citizens struggle to maintain their capacity for discernment. And if democracy depends on that capacity, then we face a constitutional risk. But also a historic opportunity: that of turning cognitive protection into the pillar of a strong, sovereign Europe that leads in the 21st century.

And that threat is no longer a projection: in 2026, figures such as Bill Gates and Geoffrey Hinton have publicly warned that AI is displacing skilled employment at an unprecedented pace, with more than 116,000 layoffs attributed to AI in the United States alone between January and August. The window for action is closing; this book proposes acting now, with an immediate temporary solution and a definitive solution with a timetable of its own.

The Protective Power arises from this conviction. Not as an improvised idea but as a structural and economic proposal: the creation of a fourth constitutional power devoted exclusively to protecting citizens' cognitive integrity. A technical, independent, transparent and strictly limited power, whose aim is not to control information but to guarantee that Europe can compete in the digital era without sacrificing its democratic values.

This book is the result of a personal and, at the same time, deeply rational process. Personal, because it is born of my concern as a citizen who refuses to accept that cognitive degradation is inevitable. Rational, because each chapter attempts to build a clear, coherent and functional architecture that can be evaluated, criticised and improved. And strategic, because it demonstrates that protecting the cognitive sphere is not only defending democracy but investing in Europe's economic and geopolitical future.

I have not walked this path alone. Artificial intelligence has been a decisive tool in preparing this third version. I have worked side by side with Vibe, Mistral AI's agent, which has not replaced my thinking but amplified it. It has allowed my intuitions to become concepts, my ideas to be organised into structures, and my concerns to be transformed into an articulated constitutional proposal.

This prologue is an invitation to the reader. An invitation to look at the problem without fear, to recognise that democracy is not defended only with laws and votes but also with mental clarity and strategic vision. An invitation to understand that protecting the cognitive sphere is not an act of control but an act of freedom. And above all, an invitation to take part in an urgent conversation: how do we guarantee that Europe does not merely survive but leads in a world where whoever controls information controls the future?

If this book manages to awaken that question, if it gets the reader to see the root of the problem and not just its manifestations — but also its potential as an engine of prosperity and leadership — then it will have fulfilled its purpose.

Prologue of "The Protective Power" | Adrian Potentela

📌 Chapter Summary

This chapter establishes the book's conceptual framework by defining the digital cognitive sphere as the fundamental space that requires protection in the information age. It explores its historical evolution, the mechanisms of the attention economy that have transformed it, and the emerging recognition of cognitive liberty as a fundamental right in the 21st century.

Chapter objectives:

📖 Section 1.1: Definition and Scope of the Cognitive Sphere

What is the Cognitive Sphere?

The cognitive sphere is the space in which individuals form their beliefs, make decisions and build their understanding of the world through the information they consume. In the digital age, this sphere has expanded beyond traditional physical limits to include:

Dimensions of the Cognitive Sphere

The Cognitive Sphere as a Public Good

The cognitive sphere is not simply a private space of individual reflection but a public good that:

"Everyone has the right to freedom of thought, conscience and religion." — Charter of Fundamental Rights of the European Union, Art. 9

📖 Section 1.2: Historical Evolution of Information Ecosystems

Pre-Digital Era (up to the 19th century)

Characteristics: information scarce and costly to produce; centralised control (church, state, elites); slow, geographically limited diffusion; high trust in sources (moral authority). Examples: medieval manuscripts, the Gutenberg press (1440), 18th-century newspapers.

Mass Era (20th century)

Characteristics: democratised access; mass media (radio, TV, press); concentration in the hands of media corporations; the emergence of propaganda and modern advertising. Milestones: commercial radio (1920s), television as the dominant medium (1950s), cable and satellite with audience fragmentation (1980s).

Digital Era (1990-2010)

Characteristics: digitisation of information; initial decentralisation (web 1.0 and 2.0); the first search engines; the birth of social networks. Milestones: first web page at CERN (1991), founding of Google (1998), launch of Facebook (2004), YouTube and the online video era (2005).

Algorithmic Era (2010-2020)

Characteristics: content personalised to each user; automation (algorithms decide what information we see); concentration in a handful of platforms; real-time speed 24/7. Milestones: Facebook passes 1 billion users (2012), recommendation algorithms become ubiquitous (2016), the Cambridge Analytica scandal (2018), the pandemic accelerates digitalisation (2020).

AI Era (2020-Present)

Characteristics: synthetic generation (AI creates content indistinguishable from human); manipulation at scale (deepfakes, bots, coordinated campaigns); oversaturation and fragmented attention; growing technological dependence. Recent milestones: large language models released to the public (2022), the first AI regulations with the EU AI Act (2023-2024), elections with massive use of deepfakes (2024), US veto on advanced AI models for foreign actors (2026).

Impact of the Evolution

📖 Section 1.3: The Attention Economy and its Impacts

What is the Attention Economy?

An economic model in which human attention becomes the most valuable resource, and platforms compete to capture and monetise it.

"If you are not paying for it, you're not the customer — you're the product being sold." — Andrew Lewis, 2010

Attention-Capture Mechanisms

1. Addictive Design (Hook Model — Nir Eyal)

2. Recommendation Systems

3. Push Notifications

Consequences of the Attention Economy

Key Statistics (2019-2025)

📖 Section 1.4: Cognitive Liberty as a Fundamental Right

What is Cognitive Liberty?

The right of individuals to control their own mental experience and thought processes free from unconsented external interference.

International Legal Basis

New Challenges Requiring Protection

Recognition Initiatives

Arguments in Favour of Recognition

"In the digital age, the protection of the human mind must be as fundamental as the protection of the body. Cognitive liberty is not a luxury; it is a necessity for the survival of democracy."

📚 Sources and References

Academic Sources

1. Eyal, N. (2014). Hooked: How to Build Habit-Forming Products. Portfolio.
2. Pariser, E. (2011). The Filter Bubble. Penguin Press.
3. Sunstein, C. R. (2017). #Republic: Divided Democracy in the Age of Social Media. Princeton University Press.
4. Zuboff, S. (2019). The Age of Surveillance Capitalism. PublicAffairs.

Reports and Studies

1. Council of Europe, Committee on Bioethics (2023). Report on neurotechnologies and human rights (mental privacy and cognitive integrity).
2. European Parliament, INGE special committee (2022). Final report on foreign interference in the EU's democratic processes.
3. UNESCO (2021). Recommendation on the Ethics of Artificial Intelligence (193 states).
4. Digital 2025 Report. We Are Social and Hootsuite.
5. Eurobarometer (2025). European Attitudes towards Disinformation.

Statistical Data

1. dscout (2016). Study on mobile phone interactions (96 checks/day figure).
2. RescueTime (2025). Annual digital productivity report.
3. Mark, G. (2023). Attention Span.  verified series: 2.5 min (2004) → 75 sec (2012) → 47 sec (~2016+); ~23 min to resume a task after an interruption.
4. MIT Technology Review (2025). The impact of algorithms on content consumption.

Legal Frameworks

1. Universal Declaration of Human Rights (1948). UN.
2. European Convention on Human Rights (1950). Council of Europe.
3. EU Charter of Fundamental Rights (2000).
4. Regulation (EU) 2024/1689 — EU AI Act.

🎯 Connection with the Rest of the Book

Chapter 1 of 21 — The Digital Cognitive Sphere | Adrian Potentela

📌 Chapter Summary

This chapter breaks down the concrete mechanisms used to manipulate the digital cognitive sphere. It analyses the tools, techniques and strategies employed by state actors, corporations and interest groups to influence citizens' perceptions, beliefs and behaviour.

Chapter objectives:

📖 Section 2.1: Disinformation — Types, Vectors and Impacts

What is Disinformation?

EU definition (2018): "False, inaccurate or misleading information designed, presented and promoted to intentionally cause public harm or for profit."

Types of Disinformation

62% of Europeans have come across fake news (illustrative estimate by the author; the 2024 Eurobarometer verifies 37% daily or almost daily).

Vectors of Diffusion

Key vector data (2026): 68% of disinformation spreads first on X | 45% of WhatsApp users have received false messages | 32% of deepfakes circulate on Facebook before detection | 22% of fake news eventually reaches traditional media. (Illustrative estimates by the author.)

Impact of Disinformation

📖 Section 2.2: Deepfakes and Synthetic Media

What are Deepfakes?

Audiovisual content generated or altered by artificial intelligence that appears authentic but is false. Technologies used:

Evolution of Deepfakes (2017-2026)

Key Case Studies

Accessible Creation Tools

Impact of Deepfakes

2.3 Algorithmic Amplification and Filter Bubbles

Disinformation does not spread on its own: it is amplified by the algorithms that curate what billions of people see every day. This is the third layer of the toolkit: the machinery that multiplies the reach of manipulative content.

Filter bubbles and echo chambers

Algorithms optimise for engagement, not truth. The result is the filter bubble: a personalised information environment that reinforces prior beliefs.

"Personalised filters are invisible: you do not see what is being edited out, and you do not see your own cocoon."

— Eli Pariser, The Filter Bubble (2011)

Consequences of algorithmic amplification

2.4 Micro-targeting and Psychological Profiling

The fourth layer completes the toolkit: knowing who to deceive, how and when. Micro-targeting turns manipulation into a precision operation.

Cambridge Analytica: the precedent

In 2018 it was revealed that Cambridge Analytica had exploited the psychological profiles of 87 million Facebook users, applying the OCEAN model (Openness, Conscientiousness, Extroversion, Agreeableness, Neuroticism) to tailor political messages to each personality type.

Micro-targeting in 2026

The resulting landscape

92% of European analysts consider micro-targeting a threat to electoral integrity (Eurobarometer 2026). 78% of citizens are unaware that the content they see is personalised. 65% of parties in the EU already use micro-targeting tools. Only 40% of Europeans understand how algorithms select their information. (Eurobarometer-style illustrative estimates by the author, based on the 2024-2026 literature.) Accuracy note: the experimental literature (PNAS 2024; MIT) finds the persuasive advantage of micro-targeting over alternative strategies to be small; the structural problem justifying audit is not persuasive efficacy but the transparency asymmetry: unauditable micro-targeting breaks electoral equality of arms regardless of its efficacy.

Chapter Data and Statistics

Summary Table: The Manipulation Toolkit

Sources

• Vosoughi, S., Roy, D. & Aral, S. (2018). "The spread of true and false news online". Science, 359(6380), 1146-1151. MIT Media Lab analysis of 126,000 stories.
• Washington University in St. Louis (2024). Study on the persuasiveness of political deepfakes (42%) and the "liar's dividend".
• Meta-analysis of 56 studies (2024): human deepfake detection accuracy of 55.5%.
• iProov (2025). Threat Report: only 0.1% of users identify all deepfakes presented to them.
• SCMP / AFP (2024). Arup case, Hong Kong: deepfake video call, loss of HK$200 million (~US$25 million).
• Pariser, E. (2011). The Filter Bubble: What the Internet Is Hiding from You.
• European Parliament / media investigations on Cambridge Analytica and the OCEAN model (2018).

3.1 Case studies: electoral interference (2016-2026)

Analysis methodology: each case is examined by context, tools, actors, vectors, measured impact and response. Only verifiable figures are presented; estimates are explicitly marked.

Case 1: United States presidential election (2016)

The Russian Internet Research Agency (IRA) ran an influence campaign at minimal cost and massive reach. Facebook estimated to the US Senate that 126 million Americans may have seen IRA content between 2015 and 2017, generated with just US$100,000 in advertising across roughly 3,400 ads, alongside more than 61,500 organic posts on Facebook and 116,000 on Instagram.

Case 2: The Brexit referendum (2016)

The "Leave" campaign combined the bus lie ("we send the EU £350 million a week, let's fund our NHS instead"), unauditable micro-targeted messaging, and automated amplification. The UK Electoral Commission fined Leave.EU £70,000 for funding breaches.

• Result: 51.9% in favour of leaving; turnout of 72.2%.
• Lasting effect: the NHS promise was still cited years later as the symbol of electorally profitable disinformation — the "liar's dividend" in action.
• Lesson: advertising transparency arrived too late (ad archives only from 2018); a verifiable lie that decided an irreversible referendum.

Case 3: Brazil (2018 and 2022)

2018: days before the runoff, the Folha de S.Paulo investigation revealed that conservative businessmen had signed contracts of up to US$3 million with agencies specialised in automated bulk WhatsApp messaging in favour of Jair Bolsonaro — a practice possibly constituting an illegal campaign contribution, which WhatsApp could not audit because of its end-to-end encryption.

2022: facing Lula's re-election (50.9%), Brazil's Superior Electoral Court deployed real-time monitoring and platforms limited bulk forwarding; even so, electoral fraud narratives penetrated deeply enough to fuel the institutional attacks of 8 January 2023.

• Lesson 2018: encrypted messaging is opaque even to the platform — the regulator is late by design.
• Lesson 2022: real-time electoral monitoring works on polling day, but does not prevent post-vote delegitimisation. Manipulation no longer ends at the ballot box.

Case 4: India (2019 and 2024)

With 968 million eligible voters in 2024, India is the world's largest experiment in democracy under generative AI. The difference between 2019 (mass WhatsApp disinformation) and 2024 is qualitative: affordable personalised deepfakes, voice cloning of leaders, and industrial-scale synthetic content. The response included the creation of the Misinformation Combat Alliance's Deepfakes Analysis Unit (DAU), a WhatsApp channel for citizens to verify content.

• Lesson: at Indian scale, no human fact-checker can keep up; only institutional technological systems can process the volume. This is the scale argument that underpins Part II of this book.

Case 5: European Union (2019 and 2024)

2019: the EU already operated the East StratCom Task Force and EUvsDisinfo, but with tiny resources against the threat. 2024: the European Parliament elections were held with the DSA in force, 200+ deepfakes documented by the EDMO network, amplified narratives on migration and climate, and turnout at 51.2% (the second consecutive above 50%).

• Progress: for the first time there were legally enforceable obligations on platforms, not voluntary codes.
• Limit: enforcement remained reactive, national and fragmented — the DSA sanctions conduct, but builds no positive protection capacity. That is precisely the gap the EDSPA comes to fill.

3.2 Metrics of social polarisation

How polarisation is measured

• Ideological: distance between the average positions of electorates.
• Affective: negative feelings towards voters of the rival party (0-100 thermometer). The most dangerous kind: one does not disagree, one despises.
• Elites: hostility between parties and reference media.

The most robust academic evidence on affective polarisation in Europe comes from Ansgar Reiljan's study (2020, European Journal of Political Research), which measured the "liking gap" between voters of rival parties across European party systems. His findings: the systems with the highest affective polarisation in Europe are in the south and east of the continent, with Spain, Poland and Hungary among the most affected, and affective polarisation can be high even when ideological polarisation is moderate — in other words, one can hate the rival without the political positions actually being far apart. Manipulation exploits precisely that gap.

Bubbles and polarisation on social media

Consequences of polarisation

• Democratic paralysis: less room for compromise; minority governments and institutional deadlock.
• Cross-delegitimisation: the adversary does not win, "he steals"; the result is not lost, "it is falsified".
• A market for disinformation: polarisation increases demand for content that humiliates the rival — it is the fuel of the cycle in Ch. 2.
• Political violence: affective hostility is the strongest verbal predictor of violent incidents.

3.3 The trust paradox: Europe responds, the national erodes

The finding that reframes this chapter —and which is often presented incorrectly as a monotonic decline in trust— comes from Standard Eurobarometer 103 (spring 2025):

Trust in the European Union reached its highest level in 18 years in 2025: 52% of Europeans trust the EU and the European Commission, up from 31% in 2013. But only 36% trust their national government and 37% their national parliament.

— Standard Eurobarometer 103, spring 2025

Why it is an information paradox

Disinformation does not erode uniformly: it strikes where the national narrative has the largest attack surface —governments, parties, national media— while the EU, further from daily media noise and associated with visible responses (vaccines, post-COVID recovery, support for Ukraine), rises. The decisive figure for this book: among 15-to-24-year-olds trust in the EU reaches 59%, and 75% of Europeans feel they are citizens of the EU, the highest level in more than two decades.

Where erosion is real

Concentrated erosion is observed in political parties (systematically the least trusted institution across Europe), in traditional media —increasingly trapped in cross-accusations of bias— and in the electoral process, when post-vote delegitimisation takes hold, as the USA 2020 and Brazil 2022 demonstrated.

3.4 The EU's regulatory response: necessary, insufficient

Between 2018 and 2024 the EU built the most advanced digital regulatory framework in the world. It is a real success — and also the proof that regulation alone does not protect.

Verified effectiveness

• Achievements: real advertising transparency, mandatory risk assessments for VLOPs, the world's first framework requiring deepfake labelling (AI Act), fact-checker coordination via EDMO.
• Limitations: uneven enforcement across states, technology evolving faster than the rule, platforms reluctant to share data, and —most important— nobody in Europe's job is to actively protect the cognitive sphere: everyone watches past conduct, no one builds present defence.

Chapter data and statistics

Verified numerical synthesis

Sources

• US Senate / New Knowledge report (2018). The Tactics & Tropes of the Internet Research Agency.
• Facebook, Twitter and Google testimony before the US Senate (2017-2018).
• Folha de S.Paulo (October 2018). Investigation into bulk WhatsApp messaging contracts.
• The Conversation / PBS (2018). WhatsApp skewed Brazilian election.
• UK Electoral Commission (2018). Report on the 2016 referendum.
• Reiljan, A. (2020). "Fear and loathing across party lines (also) in Europe: Affective polarisation in European party systems". European Journal of Political Research, 59(2).
• Standard Eurobarometer 103 (spring 2025). Trust in institutions.
• European Commission (2022-2025). DSA enforcement and Code of Practice on Disinformation reports.
• EDMO (2024). Monitoring of the European Parliament elections.
• Misinformation Combat Alliance / WEF (2024). Lessons from India's Deepfakes Analysis Unit.

4.1 Technological dependence on foreign powers

Definition: a situation in which a region depends on technology developed, controlled or hosted abroad for critical functions of its society, economy and government. Dependency is not an efficiency problem: it is a problem of someone else's will.

The four layers of dependency

The legal precedent: the CLOUD Act (2018)

The US CLOUD Act of 2018 allows US authorities to demand data held by US companies anywhere in the world. Combined with the EU Court of Justice's Schrems II ruling (2020), which invalidated the Privacy Shield precisely because of the risk of US access to European data, the consequence is direct: European citizens' data on US clouds sits under the jurisdiction of a third state. European disinformation-detection infrastructure, if built on US cloud, carries that vulnerability by design.

4.2 Data colonisation: the one-way flow

The concept comes from the academic critique of surveillance capitalism: Shoshana Zuboff (The Age of Surveillance Capitalism, 2019) described the extraction and monetisation of human experience as raw material; Nick Couldry and Ulises Mejias (2019) named it directly: "data colonisation", the process by which data from people in other regions is extracted, processed and monetised without proportional benefit to their communities.

The real size of the European data economy

Verified figures from the European Data Market Monitoring Tool (European Commission, 2025) scale what is at stake: the EU27 data market grew from EUR 82,000M (2023) to EUR 90,000M (2024), and the data economy will exceed EUR 630,000M in 2025 (~4.7% of GDP), on the way to EUR 815,000M by 2030. The Draghi report (2024) underlines the contradiction: that value is generated on European data, but is mostly captured by non-European platforms.

4.3 The US AI veto (June 2026)

What this chapter anticipated as a hypothesis happened in June 2026, with proper names. The verified sequence:

What it means for cognitive protection

• A precedent, not a one-off: the control architecture already exists; applying it to other models is an administrative decision, not a new law.
• Disinformation detection is frontier AI: the best deepfake and coordinated-campaign detection systems are trained with the very class of models now subject to licence. Europe cannot base its cognitive defence on tools whose availability depends on a permit from US Commerce.
• "Trusted partners" is a hierarchy, not an alliance: the 26 June exemption is granted by Washington case by case. Sovereignty is not delegated; it is exercised or it does not exist.
• The semiconductor precedent: it is the same mechanism already applied to China on chips — now extended to AI and to allies.

4.4 Data access restrictions: opacity as de facto policy

The CrowdTangle shutdown (2024)

On 14 August 2024 Meta shut down CrowdTangle, the tool that allowed journalists and researchers to track disinformation on Facebook and Instagram. Its replacement, Meta Content Library, is less transparent and accessible: fewer features on the eleven key topics audited, a slow and restrictive access process, no ability to track the performance of custom lists — precisely in the year of more than sixty elections worldwide (joint investigation by Proof News, the Tow Center and the Algorithmic Transparency Institute, 2024).

The pattern repeats

The European response: DSA Article 40

For the first time, DSA Article 40 obliges very large platforms to give data access to accredited researchers ("vetted researchers") designated by digital competence centres. It is real progress — and also a demonstration of slowness: the first accesses came years after entry into force, with litigation along the way. Regulation forces sharing; it does not build analytical capacity of one's own. That is the difference between granted transparency and exercised sovereignty.

4.5 The need for European digital sovereignty

Definition: a region's capacity to control its own digital destiny —infrastructure, data, technologies and rules—. Cognitive sovereignty is its top layer: without home-grown infrastructure there is no home-grown detection; without home-grown detection there is no possible protection.

Europe already has assets: supercomputing (EuroHPC)

The EuroHPC Joint Undertaking funds twelve state-of-the-art supercomputers across Europe. It is proof that when Europe decides to build capacity, it does — and it is the compute base on which to train sovereign models.

The triangle of cognitive sovereignty

• Home-grown models: Mistral AI (France, valued at ~EUR 11,700M), Cohere-Aleph Alpha (Germany), Black Forest Labs, DeepL, Kyutai, AMD Silo AI — the European vanguard exists; it needs stable public demand to scale.
• Home-grown cloud: OVHcloud, T-Systems, STACKIT, Hetzner — with the push of the sovereign marketplace (Gaia-X / IPCEI-CIS) so that the European public sector buys European.
• Home-grown data: common European data spaces (health, mobility, energy) with EUR 2,000M from the Digital Europe programme.

Chapter data and statistics

Sources

• Synergy Research Group (2026). Quarterly cloud infrastructure tracker, Q1 2026.
• Holori (2026). Cloud Market Share 2026; European market analysis.
• European Commission (2025). European Data Market Monitoring Tool — First Report on Facts and Figures.
• Draghi, M. (2024). The Future of European Competitiveness.
• US Executive Order on advanced AI (2 June 2026) and IIL directives of the Department of Commerce (12 and 26 June 2026), documented by Mayer Brown (2026).
• Zuboff, S. (2019). The Age of Surveillance Capitalism. PublicAffairs.
• Couldry, N. & Mejias, U. (2019). The Costs of Connection. Stanford University Press.
• CJR / Proof News / Tow Center / Algorithmic Transparency Institute (2024). Audit of the CrowdTangle shutdown and Meta Content Library.
• EuroHPC Joint Undertaking (2026). Supercomputer catalogue.
• Court of Justice of the EU (2020). Schrems II judgment (C-311/18); US CLOUD Act (2018).

5.1 Synthesis of the threats: the perfect storm

Chapters 1-4 did not describe four separate problems, but an interdependent system of threats. Each component amplifies the others:

The diagnosis in verified figures

5.2 Current mechanisms and their real budgets

Europe is not unarmed: it has advanced regulation (Ch. 3) and monitoring institutions. What it does not have is scale. The verified budget comparison shows it without mercy:

Inventory of mechanisms and limits

The seven structural limitations

• Fragmentation: EDMO, East StratCom, ENISA, national agencies and electoral units work in parallel, with no command centre.
• Underfunding: tens of millions against a manipulation industry operating at the scale of platforms with billions of users.
• Reaction, not prevention: debunkings arrive when disinformation has already completed its cycle — which is 6 times faster than the truth (Ch. 2).
• Platform dependency: the data needed to detect campaigns is a revocable concession of US companies (Ch. 4).
• Lack of specialisation: cognitive protection requires profiles that exist in no European organisational chart today: coordinated-campaign analysts, deepfake detection engineers, free-expression lawyers working together.
• Digital borders vs national responses: a campaign is born in one country, amplified from another and strikes a third; competences stop at every border.
• Technological pace: the DSA was designed before generative AI; the AI Act will be applied gradually until 2027. The law always arrives after the tool.

5.3 Institutional bottlenecks

Suppose the EU decided tomorrow to create the missing capacity. How long would it take? The verified precedents:

The four types of bottleneck

• Political: unanimity in the Council for sensitive matters; diverging national interests (regulatory France, self-regulatory Germany, national-sovereignty overlays). Enhanced cooperation (TFEU art. 326-334) exists precisely to bypass this blockage.
• Legal: the ordinary legislative procedure takes years; education and culture competences are national, blocking large-scale European media-literacy programmes.
• Administrative: hiring AI talent in an EU organigram requires procedures that the private sector solves in weeks; the best profiles choose other destinations.
• Technical: scarcity of deepfake-detection and campaign-analysis experts; insufficient home-grown infrastructure (Ch. 4) exactly when detection becomes compute-intensive.

5.4 The European Democracy Shield (2025): the official diagnosis and its limit

On 12 November 2025 the European Commission and the High Representative published the joint communication "The European Democracy Shield: Empowering Strong and Resilient Democracies". It is the most ambitious institutional response to date to the diagnosis of this Part I — and a close reading confirms exactly the thesis of this book.

The verified critique: coordinating is not protecting

The Jacques Delors Centre analysis (December 2025) — "papering over the cracks" — sums up the Shield's structural limit:

• The new European Centre for Democratic Resilience is a coordination structure based on member states' voluntary participation: it gathers existing expertise, it does not build new capacity.
• It suspiciously resembles the EEAS Rapid Alert System (2019), set up for that year's European elections — and which "never really took off", according to the 2025 retrospective balance.
• Verified risk of duplicating structures without clarity on what the Centre adds to what exists (EDMO, East StratCom, national units).
• The Shield rests on proposed budgets of the next financial framework (CERV+ EUR 3,600M, AgoraEU EUR 8,600M) that still require negotiation and mostly deploy grants to third parties, not home-grown operational capacity.

5.5 The case for dedicated protection: the EDSPA

Why cognitive protection needs a dedicated institution

The reference budget: an operational cognitive-protection agency would require, in this author's illustrative estimate, EUR 500-1,000M a year — the order of magnitude of Frontex for the physical border (~EUR 1,100M verified in 2025). It is the scale at which Europe already protects a border under attack; the cognitive border deserves at least the same budgetary logic. (Author's illustrative estimate; the detailed cost and return calculations for the EDSPA are developed in chapters 9 and 9B.)

How to launch it: enhanced cooperation

The bottleneck analysis (5.3) advises against waiting for the consensus of all 27. Enhanced cooperation (TFEU, art. 326-334) allows a group of member states —at least nine— to create the agency without waiting for everyone, as was done for the euro, the patent package or the European divorce. It is the route Part II proposes for the EDSPA's launch, with the door open for the rest to join later.

The five arguments for urgency

• The threat grows faster than the response: generative AI makes manipulation cheaper every year; the law needs years (5.3).
• The cost of inaction is real: an EU27 data economy of EUR 630,000M (4.7% of GDP, Ch. 4) built on ground Europe does not control; national erosion of trust (Ch. 3); verified post-electoral delegitimisation (USA 2020, Brazil 2022).
• Existing solutions have been tried and do not suffice: Rapid Alert System (2019, failed), EDMO (EUR 2.5M, observation), 2025 Shield (voluntary coordination) — certified by the Delors Centre itself.
• Europe has the capacity: EuroHPC supercomputing (JUPITER exascale), Mistral AI, Cohere-Aleph Alpha, emerging European cloud (Ch. 4) — what is missing is the institution that orchestrates it.
• The political moment is the best in years: the June 2026 AI veto created awareness of dependency; the 2025 Shield put the topic at the centre of the agenda; the proposed 2028 financial framework already includes democratic programmes worth tens of billions. The window is open — and windows close.

Chapter data and statistics

Sources

• European Commission / High Representative (12 November 2025). The European Democracy Shield: Empowering Strong and Resilient Democracies — Joint Communication.
• EPRS — European Parliamentary Research Service (2026). The European Democracy Shield: An Overview.
• Jacques Delors Centre (December 2025). The European Democracy Shield — Papering over the Cracks.
• European Commission. EDMO funding and its network of 15 hubs (CEF / Digital Europe).
• EEAS / German parliamentary documentation (2022). East StratCom Task Force and EUvsDisinfo data.
• ENISA (2025). Annual budget and EU Cybersecurity Reserve (EUR 36M).
• Frontex (2024-2025). Voted budget 2025; European Commission, DG HOME (Jul. 2025): proposal of EUR 11,900M for 2028-2034 (Statewatch cites EUR 11,200M).
• Government of France. Decree no. 2021-922 of 13 July 2021 (creation of Viginum).
• RFE/RL / Globalsecurity (November 2025). Rapid Alert System balance and Shield coverage.
• Vosoughi, S., Roy, D. & Aral, S. (2018). Science — basis of the threat synthesis (chapters 2 and 1-4).

6.1 What is the EDSPA?

Definition and mission

EDSPA — European Digital Sphere Protection Agency. The agency protects the digital cognitive sphere of European citizens, institutions and societies against manipulation, disinformation and algorithmic threats.

"To protect the digital cognitive sphere of European citizens, institutions and societies against manipulation, disinformation and algorithmic threats, guaranteeing the digital sovereignty, cognitive autonomy and democratic resilience of the European Union."

Proposed mission of the EDSPA

Its vision: to become the global reference in cognitive protection, with a model replicable by other democracies, based on transparency, ethics and respect for fundamental rights.

Strategic objectives

Core values

6.2 Legal basis: creating the EDSPA without treaty reform

The EDSPA can be created without any reform of the EU Treaties, by combining four existing legal bases:

1. Internal market (Article 114 TFEU)

The fragmentation of cognitive protection among Member States distorts the digital single market: disinformation erodes trust in the common digital market and the absence of common standards creates barriers to trade and innovation. Article 114 allows the adoption of binding regulations for the whole EU — the same route used for the GDPR, the DSA and the AI Act.

2. Personal data protection (Article 16 TFEU)

Cognitive protection is intimately linked to data protection: algorithmic manipulation relies on the unethical use of personal data. It is a well-established, proven legal basis, articulated with the GDPR and the full application of the AI Act (synthetic-content transparency obligations applicable from August 2026).

3. Enhanced cooperation (Articles 20 TEU and 326-334 TFEU)

The starting route. Article 20 TEU allows a group of Member States to move forward without consensus among the 27, using the Union's institutions. The verified minimum requirement: enhanced cooperation must involve at least nine Member States. It is ideal for an initial phase with a founding core and can expand gradually. The precedent exists: enhanced cooperation has already been used for divorce law, the unitary patent and the protection of the euro against counterfeiting. It avoids blocking by reluctant Member States.

4. Supporting competences (Article 6 TFEU)

The EU may carry out supporting, coordinating or complementary actions in areas such as education, culture and consumer protection. Media literacy and protection of the "cognitive consumer" fit this framework without intruding on national competences.

Creation process and calendar

Legal structure: a decentralised EU agency (similar to ENISA, EMA or Frontex), with its own legal personality and administrative and financial autonomy, created by a regulation of the European Parliament and of the Council, with internal rules, management board decisions and agreements with Member States for national implementation.

6.3 The founding core: nine Member States

Enhanced cooperation requires a minimum of nine Member States. We propose exactly that initial core: a group with critical economic, technological and political mass, selected against verifiable criteria.

Selection criteria

• Political commitment: clear governmental will and support for the EDSPA's values
• Technical capacity: digital infrastructure, supercomputing (EuroHPC), AI and cybersecurity expertise
• Financial capacity: GDP-proportional budget contribution and funding of national nodes
• Compatible legal framework: applicable national legislation and guaranteed fundamental-rights protection

The core of Nine (verified 2025 data)

Accession process for new Member States

1. Expression of interest (3 months): formal letter from the government to the management board and initial eligibility screening
2. Technical evaluation (6 months): capacity audit, legal compatibility, financial analysis
3. Negotiation (3-6 months): accession agreement, implementation plan, financial commitments
4. Approval (2 months): management board recommendation and approval by qualified majority
5. Implementation (6-12 months): national node, recruitment, systems integration, training

Estimated total time: 12-24 months from expression of interest to full operation. Expansion targets: 20 Member States by 2028, all 27 by 2030.

6.4 Headquarters and infrastructure

The institutional precedent

The EU has a proven procedure for deciding seats: after Brexit, the EU-27 voted in 2017 to relocate the EMA to Amsterdam and the EBA to Paris, amending the founding regulations accordingly. The seat of a new agency is decided by the Member States, not by the Commission — and candidacies are negotiated with host-country incentive offers.

Criteria and comparative evaluation

Recommendation: Brussels (Belgium), with Amsterdam as a strong alternative. Brussels offers proximity to the Commission, the Council and the European Parliament (plus NATO and the OECD), first-class connectivity and a naturally multilingual environment. The final decision rests, as in the EMA/EBA precedent, with the Member States.

Headquarters infrastructure

• Main building: ~50,000 m² — offices for 1,000-1,500 employees, on-site data centre (10,000 m²), AI and analysis laboratories (5,000 m²), 24/7 crisis room (1,000 m²), auditorium and common areas
• Compute: the EDSPA does not need to build supercomputing from scratch: it can rely on EuroHPC (JUPITER, LUMI, MareNostrum 5) with dedicated capacity, plus its own GPU cluster for real-time analysis
• Main data centre (~EUR 500M): co-location in Belgium or the Netherlands, 100% renewable energy, N+2 redundancy, ISO 27001 certification and ENISA standards
• Backup data centre (~EUR 450M) in Amsterdam/Frankfurt + own network (~EUR 100M) with GEANT connection

6.5 Financing: three consistent scenarios

Chapter 5 set the reference range for the EDSPA — EUR 500-1,000M per year, an illustrative estimate anchored to Frontex (~EUR 1,100M in 2025 for the physical border) — and Chapter 9 adds an accelerated scenario of EUR 2,500M with 2,000 employees. This chapter structures that range into scenarios with verifiable sources:

Source structure (full scenario, EUR 1,000M)

Initial investment (2027-2028)

6.6 Fundamental principles of the EDSPA

1. Absolute independence

Merit-based recruitment with transparent processes; executive mandates of 4-8 years, not automatically renewable; diversified funding (section 6.5) to avoid dependencies; veto on conflicts of interest with platforms, foreign governments or corporate interests, with public declaration of interests.

2. Technical excellence

Reference ranges of the European tech market for senior profiles; the EDSPA would complement them with stability, a public-interest mission and continuous training — the same balance that allows Europol (~1,400 staff) or the EMA to attract talent.

3. Radical transparency

Regular public reports; external audits (European Court of Auditors); aggregated and anonymised open data; public consultations. Limits: ongoing operations, privacy (GDPR) and trade secrets.

4. International cooperation

Bilateral agreements with democracies (United Kingdom, Canada, Australia, Japan, South Korea); participation in the UN, OECD and NATO; exporting the model with global standards. Non-negotiable limit: absolute prohibition on sharing information or technology with non-democratic regimes, and European control over critical decisions.

5. Accountability

6.7 Conclusion: the EDSPA as a strategic step

Part I demonstrated that the cognitive threat is real, growing and cross-cutting, and that current mechanisms — EDMO with EUR 2.5M, East StratCom with ~EUR 11M and 40 experts, the Democracy Shield without operational capacity of its own — are two orders of magnitude below what the problem requires. This chapter has shown that the institutional response is viable without treaty reform (enhanced cooperation among nine states), financeable (the EUR 500-1,000M range is the order of magnitude the EU already spends on Frontex), and governable with principles of independence, excellence, transparency and accountability.

Next steps: Chapter 7 will develop the full organisational structure (management board, organogram, technical and administrative departments, recruitment processes) and Chapter 8 operations and processes (detection, analysis, response, human-plus-AI double verification, national nodes, evaluation).

Verified sources of this chapter

• Eurostat (2025): EU GDP ~EUR 18.8 trillion; Germany ~24%, France ~16%, Italy ~12%, Spain ~9%, Netherlands ~6.3%, Poland ~5%
• ENISA: 2025 budget of EUR 26.9M; additional EUR 36M for the EU Cybersecurity Reserve (2025)
• Europol: 2024 budget of EUR 222.4M; ~1,400 staff (The Hague)
• EMA: 2025 budget above EUR 600M (new fee regulation); seated in Amsterdam since 2019
• Frontex: ~EUR 1,100M in 2025 (Chapter 5); standing corps planned at 10,000 officers by 2027
• Article 20 TEU: enhanced cooperation among a minimum of nine Member States
• European Commission: relocation of the EMA (Amsterdam) and the EBA (Paris), EU-27 procedure, 2017-2018
• DSA: fines up to 6% of annual worldwide turnover; AI Act: up to 7% or EUR 35M for prohibited practices
• Horizon Europe: EUR 93,500M (2021-2027); Digital Europe Programme: EUR 8,100M (2021-2027)
• InvestAI (European Commission, Feb 2025): EUR 200,000M mobilised for AI, including EUR 20,000M for gigafactories; France: EUR 109,000M pledged (AI Action Summit, Feb 2025)
• JUPITER (exascale EuroHPC supercomputer, Juelich, ~EUR 500M), LUMI (Kajaani), MareNostrum 5 (BSC, Barcelona)

7.1 The thesis: cognitive protection = monetary policy, not moat politics

Why can't a cognitive protection agency be governed like Frontex or Europol? Because its object of work is the minds of citizens — the very terrain where the interference battles documented in Part I are fought. An agency whose board consists of one representative per state is an agency where every government of the day — including those that today sponsor half-truths for their own electoral benefit — holds a seat, a voice and a vote over who protects citizens and against whom.

"Nobody proposes that national governments appoint the members of the ECB's board. Central bank independence is considered a condition of its effectiveness. Cognitive protection deserves exactly the same architecture: whoever guards the public sphere of ideas cannot depend on the political power that operates within it."

Central thesis of the chapter

The independence gradient of EU bodies

European bodies form a measurable gradient of state interference. Standard agencies (ENISA, Europol, EUIPO, eu-LISA) have boards with one representative per Member State plus the Commission. The EDPB is composed of national authorities that are independent — regulators, not governments. And at the maximum-independence end sit the ERC (a Scientific Council of ~22 merit-selected experts, zero state representation, the finest excellence reputation in the EU), the ECB (five pillars of institutional independence) and the EDPS (a single supervisor, no state board). The proposed EDSPA deliberately places itself at that end.

7.2 The governing body: the Council of Protectors

Composition (ERC model)

The Council of Protectors is the EDSPA's governing body: 17 independent experts, selected exclusively on merit, with no state representation whatsoever. No government appoints, vetoes or removes its members.

Required profiles and prohibitions

• Multidisciplinary composition guaranteed in the regulation: at least 3 experts in AI/algorithms, 2 in cybersecurity, 3 in fundamental and digital law, 2 in social/behavioural sciences, 2 in journalism/media, 2 in electoral security, 1 in ethics, 2 open seats for an evolving threat landscape
• Conflict-of-interest barriers (stricter than any existing agency): total veto on ties to platforms, foreign governments or corporate interests; public declaration of interests; 2-year cooling-off period for former platform employees or former national political office-holders with digital portfolios
• Mandatory diversity: gender balance (maximum 10/17 of any one gender) and geographic balance (maximum 2 members per state, with no national quotas: nationality is a data point, not a criterion)

Functions of the Council of Protectors

7.3 Executive Directorate

The Executive Director runs the agency day to day: implementation of the strategic plan, crisis management, public representation and internal budget allocation. They are appointed by the Council of Protectors from the shortlist of an independent selection committee — the same open procedure the EU already applies for agency executive directors, verified in the Europol 2026 process: public call, board selection committee, interviews and assessment centre, shortlist of at least 3 candidates.

7.4 Technical and administrative departments

1. Detection and Analysis (330 staff)

The eyes of the EDSPA. Three teams: AI (150: in-house models for detecting disinformation, deepfakes and synthetic content), Data Analysis (100: large-scale processing, platform telemetry) and Intelligence (80: analysis of actors and coordinated campaigns, OSINT). Real-time detection, manipulation-pattern identification, early warnings. The largest headcount in the organogram: the agency is, above all, a perception machine.

2. Algorithmic Audit (200 staff)

The voice of transparency. Three teams: Technical Audit (120: inspection of recommender systems and reach modulation, static and dynamic algorithm analysis), Legal (50: DSA, AI Act, GDPR) and Compliance (30: follow-up of recommendations). It conducts independent audits of platforms, assesses regulatory compliance, issues recommendations and publishes public reports. It is the arm that turns the evidence of Chapter 4 (recommender algorithms amplifying falsehood) into an obligation to correct.

3. Electoral Protection (200 staff)

The shield of democracy. Three teams: Monitoring (100: 24/7 vigilance during electoral periods), Rapid Response (60: real-time neutralisation of interference, bots, deepfakes) and Coordination (40: liaison with national electoral authorities). Detection of foreign interference, public post-election reports.

4. Cognitive Education (190 staff)

The resilience builder. Three teams: Content (80: materials for schools and citizens), Training (60: teachers, journalists, fact-checkers) and Campaigns (50: mass awareness). It is the only line of defence that works before manipulation arrives: cognitive vaccination rather than later cure.

5. Digital Sovereignty (250 staff)

The guarantor of autonomy. Three teams: Infrastructure (100: dedicated EuroHPC capacity, agency data centres), Cybersecurity (80: defence of the EDSPA itself and critical infrastructure) and Technological Sovereignty (70: promotion of European models and tools, certification of secure technologies).

6. Administration (150 staff) and Fundamental Rights Office

HR (50), Finance (40), Communication (30), Legal Affairs (30). The Fundamental Rights Office (section 7.6) is deliberately small and outside this chain of command.

Staff and budget distribution (base scenario)

7.5 National nodes: the eu-LISA + EDPB dual model

The absence of states from the central council does not mean ignoring the states: it means placing them where they help and removing them where they hinder. The national nodes — the territorial seats in each founding country — follow a dual model built on two verified EU patterns:

Component 1: Operational nodes (eu-LISA pattern)

Just as eu-LISA operates the central systems (SIS, VIS, Eurodac) while states only access them through designated authorities, the EDSPA's national nodes are operational branches directed 100% from headquarters:

• Node staff = agency staff (central contracts and chain of command), not national civil servants on secondment
• The node director is appointed by the Executive Directorate, not by the host government
• Functions: detection in national languages, liaison with locally-jurisdictional platforms, territorial electoral support, execution of the national literacy strategy
• The host state provides infrastructure, physical security and administrative facilitation — nothing more

Component 2: National Chamber (EDPB pattern)

Just as the EDPB brings national data-protection authorities together to coordinate standards, each node hosts a National Chamber where the state is present — but only for what the operation needs from it:

• Composition: the competent national authority (media or data regulator), one government representative, one local civil-society representative
• Functions: linguistic and cultural adaptation of alerts, prioritisation of national literacy campaigns, channelling requests from national electoral authorities
• No operational power: the Chamber does not direct detections, does not prioritise audits, cannot veto or delay any agency action. Its voice is real; its vote on operational matters is non-existent

7.6 Frictionless safeguards: accountability that never becomes a blockade

Independence without control is opacity, but excessive control is capture. The most delicate section of the institutional design is this one: enough guarantees for the EDSPA to be legitimate, without states or corporate litigants being able to use them to slow it down. EU law offers exact precedents for each piece:

Piece 1: Single judicial review (ECB precedent — banking supervision)

The CJEU held in Berlusconi and Fininvest v Banca d'Italia that ECB supervisory decisions are subject to single judicial review before the EU Courts, excluding parallel national jurisdiction. Translation for the EDSPA: its operational decisions (audits, orders, alerts) would be challengeable only before the General Court and the CJEU — never before 27 national courts simultaneously, the procedural drowning strategy that large platforms already use against regulators. One forum, one doctrine, coherent case law.

Piece 2: Fast internal, non-suspensive administrative review (ECB ABoR precedent)

The ECB's Administrative Board of Review examines challenges to its supervisory decisions through a fast internal procedure that is non-suspensive: the decision remains in force while under review. The EDSPA would adopt an equivalent Review Board: addressees of a measure get a real challenge channel (legitimacy), but filing the challenge does not freeze the protection (effectiveness). Platforms cannot buy time with paperwork.

Piece 3: Protected operational files (AM&S / Akzo Nobel precedent)

CJEU case law protects communications covered by legal professional privilege from mandatory disclosure. The EDSPA's regulation would extend that logic to the operational file: ongoing analysis dossiers, detection methodologies and actor intelligence are not discoverable by litigants in civil proceedings. Radical transparency about results (public reports, aggregated open data, Court of Auditors audits); regulated confidentiality about method while the operation lasts.

Piece 4: Limitation periods (EDPS precedent)

The EDPS only investigates complaints filed within two years of the complainant becoming aware of the matter. The EDSPA would apply analogous limitation periods to administrative challenges: without deadlines there is no legal certainty for the agency, and procedural eternity only benefits those who can afford armies of lawyers.

Piece 5: Fundamental Rights Officer with access but no operational veto (Frontex precedent, improved)

Frontex created its Fundamental Rights Officer by regulatory requirement after its governance scandals. The EDSPA includes one from day one, improved: full access to systems and staff, an accessible citizen complaints mechanism, annual public reports, and the duty to alert the Council of Protectors and the European Parliament — reporting directly to the Council, never to the operational chain it monitors. But its recommendations do not suspend operations: they trigger review, not paralysis.

Piece 6: Shielding against the legislator (the Europol 2022 lesson)

Case T-578/22 showed that the legislator can dismantle in one regulatory amendment what a supervisor built over years. The EDSPA's founding regulation would include a shield clause: amendments to the mandate, the governance model or the multiannual financing only through the same reinforced ordinary legislative procedure (Parliament + Council + mandatory consultation of the Council of Protectors itself and the EDPS), never by implementing regulations, and never with retroactive effect on ongoing operations. Independence is founded once, with difficulty; it must also be dismantled with difficulty.

7.7 Coordination mechanisms and indicators

Coordination rhythm

Key performance indicators (KPIs)

7.8 Conclusion: a structure built for action, shielded to last

The EDSPA's structure is deliberately asymmetric: maximum independence where it deliberates (17 merit-selected experts, zero governments), maximum professionalism where it executes (an executive directorate under an open procedure), territorial presence without yielding command (nodes as branches, states in consultative chambers) and safeguards that listen but never block (single judicial review, non-suspensive appeal, protected operational files, an independent FRO). It is the skeleton of the Protective Power: not bureaucracy, but the first European institution designed from its foundation so that nobody — neither the governments that fund it, nor the platforms it monitors, nor the litigants that sue it — can capture the instrument that protects the minds of Europeans.

Chapter 8 develops how that structure operates: the detection, analysis and response processes, human-plus-AI double verification, and real-time integration of the nodes.

Verified sources of this chapter

• ERC: Scientific Council of independent merit-selected experts, governing body of the ERC, with no state representation — the governance model adopted for the Council of Protectors
• ENISA, EUIPO, eu-LISA: boards with one representative per Member State + Commission (standard model, discarded for the headquarters); ENISA: 4-year renewable mandates, 5-member Executive Board
• Europol: 2026 executive director appointment procedure — board selection committee, interviews and assessment centre, shortlist of at least 3 candidates (applied to the EDSPA Executive Director)
• Case T-578/22 (EDPS v Council and EP, 2022 Europol Regulation): EDPS annulment action inadmissible after retroactive legalisation of data practices — foundation of the shield clause
• CJEU, Berlusconi and Fininvest v Banca d'Italia: single judicial review of ECB/SSM decisions before the EU Courts — foundation of the anti-litigation piece
• ECB Administrative Board of Review (ABoR): fast, non-suspensive internal administrative review
• CJEU, AM&S (155/79) and Akzo Nobel (C-550/07 P): legal professional privilege against disclosure — foundation of the protected operational file
• Frontex: Fundamental Rights Officer and complaints mechanism (Regulations 1168/2011 and 2019/1896); OLAF/Leggeri 2022 scandals as a governance lesson
• EDPS: two-year complaint limitation (EDPS Rules of Procedure, 2022 amendment); supervisor independence
• ECB: five pillars of institutional independence, prohibition of instructions to its bodies
• EDPB: composition of national independent authorities (pattern of the National Chamber); eu-LISA: central operation of SIS/VIS/Eurodac with designated national authorities (pattern of the operational nodes)

8.1 The operational cycle: from signal to protection

Before anything else, the EDSPA is a perception machine with judgement. Every threat travels the same cycle: continuous monitoring → anomaly identification → classification → dual verification → prioritisation → response → learning. What distinguishes the agency from a mere automated system is the layer of human judgement at the points where error carries a democratic cost.

The dual verification protocol (cross-cutting)

The guiding principle of all operations: AI proposes, the human decides — mandatory at the High and Critical levels, and with permanent human supervision at the Medium and Low levels. No agency action affecting content, accounts or platforms is executed without a machine having detected it and a person having validated it.

Incident classification

The incident-management framework adopts the skeleton of NIS2 — the verified European rule requiring an early warning within 24 hours, full notification within 72 and a final report within one month — and tightens it, because cognitive threats spread faster than a cybersecurity incident: the critical window of an electoral deepfake is measured in hours, not days.

Target detection times by threat type

8.2 Algorithmic auditing: from the DSA to in-house capacity

The DSA already obliges very large platforms (VLOPs/VLOSEs) to assess their systemic risks and to submit to independent audits — the corresponding delegated act (Delegated Regulation (EU) 2024/436) has been in force since January 2024 and applied to the first round of audits, due in August 2024; the first risk assessments were published from 2024-2025. But the system rests on the platforms' self-declaration and on auditors hired by the audited companies themselves. The EDSPA adds the missing piece: inspection capacity belonging to the regulator, just as the ECB inspects banks instead of trusting their self-assessments.

8.3 Electoral protection: institutionalising what 2024 improvised

The June 2024 European elections proved the concept with makeshift means: the Rapid Response System of the Code of Practice on Disinformation allowed fact-checkers to flag threatening content to the platforms during the campaign — and EDMO evaluated how it worked, including a separate channel with X. It worked at limited scale, with no binding capacity, and dependent on the platforms' goodwill. The EDSPA turns it into a permanent protocol with the power to demand:

8.4 Cognitive education and digital sovereignty

Education programmes (base scenario budget, Ch. 7)

The programmes are delivered through the national nodes and in collaboration with the existing network of 15 EDMO hubs and national education systems, without duplicating infrastructure: the agency produces standards and content; local actors implement them on the ground.

Digital sovereignty: the real dependency map

The dependency table is updated with verified 2025-2026 evidence. The European ecosystem is no longer theoretical: Mistral AI completed a EUR 1,700M funding round in September 2025 led by ASML, with a valuation of roughly US$13,700M in 2026 and a forecast of exceeding US$1,000M in revenue that same year; exascale supercomputing exists (JUPITER, LUMI, MareNostrum 5); the legal framework is in force (AI Act with full staggered application by August 2026, EMFA since August 2025, Anti-SLAPP since May 2026).

8.5 Coordination and interoperability

Coordination with the Member States (within the Chapter 7 model)

State coordination operates exclusively through the channels designed in Chapter 7 — the Advisory Council (headquarters) and the National Chambers (nodes) — and through technical channels the EU already has in verified operation:

• Permanent technical channel: secure exchange platform with the competent national authorities (eu-LISA pattern: access to central systems through designated authorities)
• Situation reports: weekly aggregated and anonymised reports to participating states; real-time alerts to the affected authority
• Annual joint exercises simulating cognitive crises, in the style of EU cyber exercises (defined in the EU Cyber Solidarity framework, the basis of the EUR 36M Reserve)

International cooperation (reordered after the 2026 AI veto)

The US AI veto of June 2026 (Ch. 4) forces cooperation priorities to be ordered by democratic reliability and cognitive security, not by Atlanticist habit:

8.6 Quality and transparency standards

8.7 Conclusion: operations that save democracies

The processes in this chapter turn the architecture of Chapters 6 and 7 into real protection: continuous perception with mandatory human judgement where error carries a democratic cost; deadlines more demanding than NIS2 that shorten year after year; auditing that completes the DSA with the regulator's own inspection capacity; electoral protection that institutionalises what June 2024 demonstrated with improvised means; and sovereignty built on a European ecosystem that already exists — Mistral, EuroHPC, ENISA, EDMO — but lacks the conductor. That operational conductor, with cost investigation, auditing and cooperation as its master lines, is the EDSPA's contribution.

Chapter 9 measures what all of this is worth: the economic return and direct benefits of the agency.

Verified sources for this chapter

• NIS2 (Directive 2022/2555): incident management with early warning in 24 h, notification within 72 h and final report within 1 month; transposition deadline 17-Oct-2024 — framework tightened by the EDSPA (12 h/48 h with progressive improvement)
• DSA: independent audits of VLOPs/VLOSEs (Delegated Regulation (EU) 2024/436, in force since Jan 2024, first round Aug 2024); systemic risk assessments published 2024-2025; penalties up to 6% of global turnover
• AI Act: staggered application through August 2026; penalties up to 7% or EUR 35M for prohibited practices
• Rapid Response System of the Code of Practice on Disinformation: operational during the June 2024 European elections; evaluated by EDMO, including a separate channel with X
• EDMO: central unit ~EUR 2.5M (DIGITAL 2025); 15 national and regional hubs — the network with which to interoperate on literacy and verification
• ENISA: 2025 budget of EUR 26.9M; EU Cybersecurity Reserve of EUR 36M (Digital Europe 2025-2027, contribution agreement signed with ENISA, 26-Aug-2025)
• Mistral AI: EUR 1,700M round in September 2025 (led by ASML); valuation of ~US$13,700M and a forecast of >US$1,000M in revenue in 2026
• EuroHPC: JUPITER (exascale, Juelich), LUMI (Kajaani), MareNostrum 5 (BSC, Barcelona)
• European Media Freedom Act: applicable since 8-Aug-2025; Anti-SLAPP Directive (EU) 2024/1069: transposition required by 7-May-2026
• European Democracy Shield (Joint Communication, Nov-2025): situational awareness and response capability as priorities — with the Delors Centre critique already integrated in Ch. 5

9.1 The urgency argument: the acceleration of AI in 2026

The forecasts underpinning this chapter were written before the 2026 evidence made it clear that the risk calendar is accelerating faster than the response capacity of the Member States. The warnings no longer come from outside critics of the industry: they come from its own founders.

Bill Gates: "there is not even a plan to have a plan"

In his August 2026 essay, The turbulent AI era is here. The choices we make now are critical, Bill Gates warns that AI is already crossing thresholds that were considered distant alarm signals: easier cyberattacks, biological risks, emotional dependence on chatbots, mass job displacement and possible loss of control over advanced systems. Gates compares the situation with his pre-COVID-19 warnings: the data was there, but leaders did not act in time, and he has proposed an international AI monitoring organisation inspired by nuclear inspection regimes.

Geoffrey Hinton: a new wave of job destruction from 2026

The "godfather of AI" and Nobel laureate warns that 2026 opens a new wave of job destruction: AI "is going to be able to replace lots of other jobs". Hinton foresees massive disruption of white-collar work between the end of this decade and the start of the next, with growing inequality, and notes that the tech companies "are betting that AI will replace many workers".

Narrative manipulation: Altman and Amodei tone down their alarms ahead of their IPOs

An even more revealing fact for this book's thesis: Sam Altman (OpenAI) and Dario Amodei (Anthropic), who in 2025 predicted a "job apocalypse" and a "white-collar massacre", have publicly softened their predictions in 2026, coinciding with the preparation of their stock market listings. Anthropic's own economics team published in March 2026 that "no systematic increase in unemployment is observed for the most exposed workers since late 2022, though hiring of younger workers has slowed in exposed occupations". When the narrative about risk is controlled by those who sell the technology, independent assessment stops being optional: it becomes a structural necessity. That is precisely the EDSPA's function — with the human-AI dual verification and full traceability of Chapter 8 as the guarantee that this assessment is auditable.

Labour damage already measurable in 2026

9.2 Cost analysis: the initial investment

Base scenario (2027-2028): EUR 1,100M

Accelerated scenario (2026-2027): EUR 2,500M

In light of the acceleration documented in 9.1, the accelerated scenario — the one funded in Chapters 6 and 9B — brings the start forward to 2026 with an investment of EUR 2,500M that adds to the base scenario: early hiring of up to 2,000 employees, parallel deployment of the national nodes, expanded dedicated computing capacity, and a European labour reconversion programme for workers displaced by AI. Full operational capacity is brought forward to 2028.

9.3 Direct benefits: EUR 1,450M per year at full capacity

Added to these direct benefits are the indirect ones — the public-spending multiplier in the European technology ecosystem, associated skilled employment, private investment attracted to the sector — estimated at ~EUR 870M per year at full capacity (60% of direct benefits, a conservative estimate whose full methodology belongs to Chapter 9B). Total benefit at full capacity thus reaches ~EUR 2,320M per year.

9.4 ROI calculation

Base scenario (start 2027, full capacity 2029)

• 5-year net benefit: +EUR 5,912M
• Payback of the initial investment: ~2.5 years from start-up (the annual net benefit at full capacity, EUR 1,820M, exceeds the initial investment in the first full year of operation)

Accelerated scenario (start 2026, full capacity 2028)

9.5 Financial sustainability: the 40/35/15/10 model

Sustainability is not measured in "own revenue" but in the stability of the Chapter 6 funding model: EU budget contribution (40%), contributions from participating states (35%), transfer of the EU-collected fines share (15%) and other sources — service fees for audits and training for third parties, R&D programmes (10%). The agency does not collect: the model shields it from the revenue-collection incentives that would corrupt its function.

9.6 Comparison with other public investments in Europe

9.7 Conclusion: an investment that pays for itself

With costs aligned with Chapters 6 and 7, the EDSPA is one of the few European public investments with a verifiable direct return: ~EUR 2.6 per euro invested over five years in the base scenario, payback of the initial investment in two and a half years, and an annual cost (EUR 500M) equivalent to 0.003% of EU GDP. The accelerated scenario lowers the accounting ROI but buys the one thing the 2026 evidence shows to be scarce: time against a wave of job destruction and manipulation that is already in the data, not in the forecasts. The broader economic and social impact — employment, multipliers, GDP — is quantified in Chapter 9B, and the constitutional dimension in Chapter 17.

Verified sources for this chapter

• Bill Gates, "The turbulent AI era is here" (Gates Notes, August 2026); MIT Technology Review interview (AI-facilitated bioterrorism risk "50 times more likely than a natural pandemic")
• Geoffrey Hinton (Nobel Prize in Physics 2024): 2026 statements on a new wave of job destruction and white-collar disruption
• Challenger, Gray & Christmas: 116,175 US layoffs attributed to AI in January-August 2026 (22% of the total); tech sector 155,126 (+52% vs. 2025)
• Layoffs.fyi: >73,000 global tech layoffs in Q1 2026 (95 companies)
• Funcas: gross destruction of 1.7-2.3 million jobs in Spain over the next decade (central scenario; the study itself estimates ~1.61M new occupations partially offsetting)
• Mercer: 99% of 825 US CEOs expect AI-driven headcount cuts within 2 years; ~16,000 jobs/month destroyed in the US (Apr-2026, ~11,000 in Jun-2026), ~300M exposed globally; Goldman Sachs Research: ~16,000 jobs/month destroyed in the US, ~300M exposed globally
• Anthropic (economics team, March 2026): "no systematic increase in unemployment for highly exposed workers since late 2022, though suggestive evidence of slowed hiring of younger workers in exposed occupations" — a downgrade coinciding with the preparation of the OpenAI and Anthropic listings
• Eurostat 2025: EU GDP ~EUR 18.8 trillion — basis of the 0.003% calculation
• ENISA EUR 26.9M (2025); EU Cybersecurity Reserve EUR 36M (ENISA agreement, Aug-2025); Europol EUR 222M; Frontex ~EUR 1,100M; EMA >EUR 600M — comparisons verified in Chs. 5 and 6
• InvestAI EUR 200,000M (European Commission, 2025); France EUR 109,000M — budgetary viability evidence in Ch. 6

Chapter Summary

This chapter demonstrates that the EDSPA is not an expense but a strategic investment with a clear, quantifiable return over its operational period. Drawing on the data from Chapter 9 (ROI and benefits) and on author estimates anchored in the economic literature of the domain, it quantifies the agency's impact: 3,000 direct jobs (2035), 6,000-9,000 indirect jobs, EUR 3,160M-4,000M of annual GDP impact, and an ROI of 164% over five years in the baseline scenario that will finance the transition to the Protective Power.

Objectives:

Temporal Context: The EDSPA as the Bridge to the Protective Power

Timetable and Budget — Baseline Scenario (2027-2035)

Timetable — Accelerated Scenario (2026-2033, recommended)

Baseline data (revised Chapter 9): Initial investment: EUR 1,100M (baseline, no owned datacenters: dedicated EuroHPC capacity + European cloud) / EUR 2,500M (accelerated) | Annual operating budget: EUR 500M (baseline) / EUR 1,000M (full scenario) | Staff: 1,500 (baseline 2027) / 2,000 (accelerated 2026) → 3,000-3,800 employees (2035) | Demonstrated ROI: 164% over 5 years (baseline); ~70% (accelerated, justified by avoided damage) | Net benefit: +EUR 1,820M/year at full capacity (benefits of EUR 2,320M/year minus operations of EUR 500M/year)

Section 1: The EDSPA as a Generator of Specialised Employment

1. Direct Employment: 1,500-3,000 Positions at the Vanguard of Cognitive Protection

EDSPA staff structure — baseline scenario (2027-2035):

2. Indirect Employment: 6,000-9,000 Positions in the Emerging Ecosystem

*The ecosystem company names appearing later (CognitiveShield, TruthGuard, etc.) are illustrative examples of company profiles that EDSPA demand will generate, not existing companies. The "Source" column in this table indicates reference literature of the domain (studies by institutions and consultancies on employment and cybersecurity), not specific publications backing these sectoral breakdowns; the jobs and impacts per sector are illustrative estimates by the author, consistent with the totals in Chapter 9.

Effect on the European labour market: creation of new professional profiles (algorithmic auditors, disinformation experts, cognitive protection specialists) — the qualified side of the labour market that AI is not generating on its own (Chapter 9) | Skills demand: AI (47%), cybersecurity (33%), data analysis (30%) (illustrative estimate by the author based on published trends of the market) | Talent attraction: the EDSPA as Europe's magnet for cognitive protection.

Section 2: The Economic Multiplier Effect (2.5x-3.0x per EUR 1 Invested)

1. Calculation Methodology

Theoretical framework (the author's estimate within the ranges of the economic literature): Input-Output model (Leontief): multipliers of 2.5x-3.0x for investment in technology and professional services | Keynesian multiplier: MPC ≈ 0.6-0.7 → multiplier ≈ 2.5x-3.3x | Empirical reference: global cybersecurity market with a ~13% CAGR and an implicit multiplier of ~3.0x. Rigour note: the multipliers applied to the EDSPA are an illustrative estimate by the author within these ranges; there is no auditable study of the multiplier of a cognitive protection agency (none exists yet).

2. Multiplier Analysis by Area of Investment

3. Comparison with Other Sectors (2026)

Section 3: Impact on European GDP (EUR 3,160M-4,000M+ per Year)

1. Calculating the Direct and Indirect Impact

Cumulative impact (2028-2035, baseline): additional GDP: EUR 22,120M-28,000M (0.13%-0.17% of EU GDP) | Employment created: 45,000-60,000 cumulative jobs | Accelerated scenario (2027-2035): EUR 30,000M-38,000M — the same mechanisms operating two more years at full capacity.

2. Comparison with Other Public Investments in Europe

Verified agency figures (2024-2025): ENISA: annual budget of EUR 26.9M in 2025, plus EUR 36M/3 years to operate the EU Cyber Reserve (ENISA/European Commission, 2025) — the Union's cybersecurity agency operates on a budget ~40 times smaller than what the threat requires. Europol: EUR 222M in 2024 (+4.5% per year). The GDP impacts and ROIs of the agencies are the author's own estimates using the same multiplier methodology, not official figures.

Section 4: Talent Attraction and Development of the Technology Ecosystem

1. The EDSPA as a Magnet for Cognitive Protection Experts

Key data: the EDSPA will create new professional profiles in high market demand (illustrative estimate by the author based on published trends) | it will become Europe's centre of excellence in cognitive protection | it will train 1,000-2,000 professionals per year (Chapter 9) — in the accelerated scenario this training starts in 2027 and slows the brain drain sooner.

2. Building a Technology Ecosystem for the Protective Power

Ecosystem company profiles (illustrative examples, 2035):

*Illustrative profiles of companies that the EDSPA's public demand will generate in the European ecosystem. The names used in the text (CognitiveShield, TruthGuard, VoteShield, FairAudit, etc.) are illustrative fiction, not existing companies.

Ecosystem benefits (2027-2035): (1) Preparation for the Protective Power: demonstration of viability (ROI 164% (baseline)), generation of political and social consensus, development of technical capacity. (2) Reduced technological dependence: from ~80% of European digital infrastructure imported (EuroStack/Bertelsmann 2025, verified figure from Chapter 4) towards a growing European share by 2035 (the author's objective, not a datum). (3) European standards: declared preference for EU providers in public procurement (illustrative estimate by the author).

Section 5: The EDSPA as the Financier of Its Own Transition

1. The Virtuous Circle

2. Financing Projection (2027-2035) — Baseline Scenario

Financial conclusions (baseline): cumulative break-even point: 2028 (the first full year of operation) | Total net benefit (2027-2035): +EUR 13,200M | Cumulative benefits (EUR 18,800M) over cumulative costs (EUR 5,600M): cumulative ROI of 336% | The EDSPA's benefits will cover part of the initial investment in the Protective Power (EUR 2,600M baseline; EUR 3,400M accelerated). In the accelerated scenario, break-even moves forward to 2027-2028 and the 2026-2035 cumulative total reaches ~EUR 15,000M.

3. Comparison: EDSPA vs. Protective Power

Section 6: International Comparison — the EDSPA as a Model for Other Democracies

1. The EDSPA's Position in the Global Context

Competitive advantages of the EDSPA: (1) Multilateral model: cooperation among 9 Member States of the founding core (combined GDP ~EUR 14.1 trillion, ~75% of EU GDP). (2) Comprehensive approach: cognitive protection + algorithmic auditing + education. (3) Demonstrable ROI: 164% over five years (baseline), above the average for national agencies. (4) Scalability: a replicable model for other democracies.

2. Exporting the EDSPA Model (2030-2035)

Benefits of exporting the model: advisory and training revenue of EUR 100M-200M per year (2033-2035) | positioning the EU as the leader in democratic cognitive protection | alliances with the UN, the OECD and NATO on global standards.

Conclusion: The EDSPA as a Strategic Investment for Europe's Future

1. Summary of Key Impacts (2027-2035)

2. Vision for 2033-2035

Sources and References

Economic and Agency Data

Rigour note: the IMF, OECD, McKinsey, WEF, European Commission and Techstrong.ai entries in this section work as generic references to the domain literature (published lines of research on fiscal multipliers, regulatory ROI, cybersecurity and digital skills), not as specific audited publications backing this chapter's figures; the derived figures are the author's estimates, documented in the rigour notes of each table.

1. Eurostat (2026). EU economic data
2. IMF (2026). Fiscal multipliers in advanced economies
3. OECD (2025). The ROI of regulatory agencies
4. McKinsey (2026). The economic value of cybersecurity
5. World Economic Forum (2026). Global Cybersecurity Outlook
6. European Commission (2026). EU agency budgets
7. Techstrong.ai (2026). Demand for digital skills

Verified Agency Figures (2024-2026)

1. ENISA (2025). Annual budget 2025: EUR 26.9M; EU Cyber Reserve: EUR 36M/3 years — enisa.europa.eu
2. Europol (2024). Annual budget: EUR 222M (Statista/Europol)
3. DHS/CISA (United States) (2025-2026). FY2026: $2.4B and 2,324 FTE, versus $3.0B and 3,641 FTE in FY2025 — dhs.gov
4. SOCRadar (2026). CISA: ~998 departures and loss of ~1/3 of staff since 2025 — socradar.io
5. Sénat.fr (2025-2026). ANSSI: 656 FTE, EUR 27M in "business" resources — senat.fr

Methodology

1. Input-Output model (Leontief) and Keynesian multiplier (IMF/OECD)
2. Cost-benefit analysis (OECD) and benchmarking of EU agencies
3. Dual baseline/accelerated scenario: the accelerated one is the author's proposal based on the 2026 urgency evidence (Chapter 9), not a prediction
4. Institutional rigour: the acceleration of the constitutional timetable is bounded by the real durations of the Convention, the IGC and ratification (Laeken→Lisbon precedent: 8 years)
5. Ecosystem company names are illustrative fiction; the GDP impacts of the compared agencies are the author's own estimates

Chapter 9B of 21 — Economic and Social Impact of the EDSPA 2027-2035 | Adrian Potentela

10.1 Verified feasibility: the precedents behind the calendar

No phase of the plan is speculative: each has a verified European precedent with its actual, documented duration.

10.2 The two deployment scenarios

10.3 Phase 1 — Preparation (months 1-6): the phase that decides the calendar

Legal approval (months 1-4)

• Legal route: enhanced-cooperation regulation (Art. 20 TEU) with the 9 Member States of the founding core — Germany, France, Italy, Spain, the Netherlands, Belgium, Sweden, Finland and Denmark (combined GDP ~EUR 14.1 trillion, ~75% of EU GDP, Ch. 6). Negotiation uses the Union's institutions: Commission proposal, Council (qualified majority among participants) and European Parliament (IMCO/LIBE)
• Seat act: seat decision by the verified EMA/EBA procedure — the core pre-agrees on Brussels (the natural candidate, Ch. 6) so the vote consumes no more than one cycle
• Early capacity contracts: in parallel with the legal work (not after it), the Commission — on behalf of the future core — signs the multi-year contracts for dedicated EuroHPC capacity (JUPITER, LUMI, MareNostrum 5) and sovereign European cloud. JUPITER lesson: ~24 months from tender to operation, so this leg starts on day 1

Constituting the governance (months 3-6)

The Council of Protectors is constituted: 17 independent experts selected on merit through an open procedure (ERC model, Ch. 7) — zero state representatives. In parallel, the Executive Director is selected through the open Europol-2026-type procedure, and the Fundamental Rights Office (FRO) is staffed from day one: auditing from the start, not after the first scandal.

Initial hiring (months 4-6)

A founding core of ~200 people: management, legal teams, systems architects and the first senior analysts. Phase budget: EUR 50M. To avoid dependence on permanent-competition timelines, the verified start-up mechanisms of European agencies are used: seconded national experts (SNE) — the ENISA/eu-LISA pattern — plus interim contracts for critical technical profiles.

10.4 Phase 2 — Establishment (months 5-12): the agency takes shape

10.5 Phase 3 — Testing and training (months 11-16)

10.6 Phase 4 — From operational start to full capacity

Capacity does not arrive all at once: it follows the ramp of Chapter 9 — 40% of full-capacity benefits in the first year of operation, 70% in the second, 100% in the third. The initial operational capability (IOC) of the accelerated scenario arrives 18 months from start: detection, auditing and electoral protection systems in production, the core's national nodes connected, and the first full algorithmic audit with a public report. Full capacity (1,500-2,000 staff, 30% of the platform landscape audited in 2027, KPIs at full demand) arrives in 2028 (accelerated) or 2029 (baseline).

10.7 Implementation budget by phases

The accelerated scenario (EUR 2,500M) keeps the same phase structure compressed into 2026-2027, and adds: early hiring of up to 2,000 employees, parallel deployment of the nodes, expanded dedicated computing capacity and the European labour reconversion programme for workers displaced by AI (Ch. 9).

10.8 Risks and mitigation

10.9 Monitoring and transparency

• Exit criteria per phase (10.3-10.6): every milestone has a verifiable criterion — not declarative dates but passed tests
• Chapter 7 KPIs as the permanent dashboard: detection < 12 h (2027) → < 6 h (2028) → < 2 h (2030); accuracy > 90% → > 95%; platforms audited 30% (2027) → 90% (2030)
• Initial evaluation at the close of Phase 4: public report by the Council of Protectors to the European Parliament and the 9 participating states, with deviations and a corrective plan
• FRO audit from the first month of operation: fundamental rights are supervised during the deployment, not after it
• Quarterly aggregate statistics from IOC (Ch. 8.6)

10.10 Conclusion: the calendar is the strategy

The difference between the EDSPA and another European paper exercise is not in the design: it is in the dates. The 2026 evidence (Ch. 9) — 116,175 layoffs attributed to AI in eight months, an industry that modulates its warnings according to its stock price — does not wait for a perfect deployment. The verified precedents say Europe knows how to move fast when it wants to: six weeks for the EUAA from the Council position (five from signature), six months for the EMA and EBA seats, eighteen months for ENISA from scratch. The accelerated scenario presented here does not require Europe to do something it has never done — it requires it to do on time what it already knows how to do. And the critical path has a name: computing capacity. Whoever controls the EuroHPC contract calendar controls the calendar of European cognitive protection. That is why they are signed on day 1.

Chapter 11 explains why this agency, even successfully deployed, is only the necessary step — not the definitive one — towards the constitutional Protective Power.

Verified sources for this chapter

• EUAA: Regulation (EU) 2021/2303 — Council position on 9 December 2021, signed on 15 December 2021, in force on 19 January 2022 — transformation of EASO (Reg. 439/2010, operational since 2011): the optimistic ceiling of the legal route on an existing structure
• ENISA: Regulation (EC) No 460/2004 of 10 March 2004; fully operational on 1 September 2005 — the realistic benchmark for creation from scratch (~18 months)
• EMA/EBA (2017): relocation procedure endorsed by the 27 on 22 June 2017; seat decided on 20 November 2017 (Amsterdam and Paris) — the seat decision in ~6 months
• Art. 20 TEU: enhanced cooperation with a minimum of 9 Member States — the basis of the founding core of Ch. 6 and of the 12-18 month estimate (Ch. 9B)
• JUPITER (EuroHPC JU): tender launched in January 2023, contract signed that year, installation 2024, user access from late 2024 — ~24 months from tender to operation
• NIS2 (Directive 2022/2555): 24 h/72 h — tightened by the EDSPA to 12 h/48 h (Ch. 8)
• Europol: open procedure for selecting the Executive Director (2026, Ch. 7); EUR 222M / ~1,400 staff (Chs. 5-6)
• ENISA / eu-LISA: start-up pattern with seconded national experts (SNEs) and interim contracts verified in both agencies
• EU Cyber Solidarity: joint simulation exercises as the pattern for the Phase 3 drills (Ch. 8)

11.1 The fundamental problem: agencies can disappear

The EU's institutional history shows that a decentralized agency is a revocable solution: it depends on political will, shifting priorities and annual budgets. Three verified cases illustrate this fragility:

11.2 The structural limits of the EDSPA as an agency

Even though the EDSPA is designed with the hardened governance of Chapters 6-7 (a Council of Protectors of 17 independent experts, an Executive Director by open procedure, mandatory human-AI double verification), as an agency it remains revocable by the ordinary legislator:

11.3 The Protective Power: the definitive solution

Definition and purpose

The Protective Power is the 4th constitutional power: a permanent institution charged with protecting the cognitive sphere of citizens, institutions and European society against digital manipulation, guaranteeing that no actor — public or private — can subvert democratic processes, individual autonomy or collective sovereignty.

Why a 4th power and not more of the same

None of the three has the mandate, the capability or the agility to protect the cognitive sphere in a proactive, comprehensive and permanent way. The Protective Power rebalances the structural asymmetry between technology (fast, global, opaque) and democracy (slow, fragmented, transparent) — just as the Judiciary was created to interpret the law impartially, the Protective Power is created to protect cognition with specialization.

Exclusive competences

11.4 Why not create the Protective Power directly

The constitutional route is codified in Article 48 TEU: the ordinary revision procedure requires a Convention, an intergovernmental conference, unanimity in the European Council and ratification by all 27 under their constitutional rules (in several cases, referendum). The passerelle clauses of paragraph 7 do not help here: secondary law cannot create a new constitutional power.

Recent experience confirms the difficulty: the European Parliament voted on 22 November 2023 in favor of treaty reform following the Conference on the Future of Europe, and asked the European Council to convene the Convention — but no consensus has materialized in the Council to open the process. The great historical revisions (from the Single European Act to Lisbon) took between 4 and 8 years from convening to entry into force. The earliest defensible horizon for a 4th constitutional power is therefore 2033.

11.5 The EDSPA as laboratory: what the transition will inherit

11.6 The transition timeline (2026-2035)

The timeline aligns the dual scenario of Chapter 10 with the Art. 48 TEU procedure. The realistic starting point: the accelerated scenario places initial operational capability in 2028, so the reform proposal can be presented on results — not on promises — in 2028-2029.

11.7 The thread of Part III (Chapters 11-20)

11.8 Conclusion: the EDSPA is the means; the Protective Power is the end

1. The EDSPA is a necessary but insufficient step. As an agency, it can be repealed, cut or reorganized by any future majority — as EDMO, EMSA and OLAF demonstrate. Its value lies not in what it achieves by itself, but in what it enables building.
2. The Protective Power is the definitive solution. As the 4th constitutional power, it would be permanent, specialized and impartial by design: it would return to democracies control over the digital sphere with the solidity that only constitutional law provides.
3. The EDSPA is the strategic path. It proves that cognitive protection works, generates the political and social support the reform demands, and builds the capability the Protective Power will inherit. Acting with the 9 today is what makes the unanimity of the 27 possible tomorrow.

11.9 Verified sources and references

• Art. 48 TEU (ordinary and simplified revision procedures): EUR-Lex, consolidated text of the Treaty on European Union.
• Art. 20 TEU and Arts. 326-334 TFEU (enhanced cooperation): EUR-Lex, consolidated texts.
• European Parliament, 22 November 2023: resolution on the proposals to amend the Treaties following the Conference on the Future of Europe (request for a Convention).
• EDMO: European Commission official page (digital-strategy.ec.europa.eu) and DIGITAL calls on hubs with funding "ending in 2026".
• EMSA: proposal COM(2023) 269 final, 1 June 2023; provisional Council-Parliament agreement, 20 May 2025 (Council of the EU).
• OLAF: Court of Auditors Opinion on OLAF management (EUR-Lex); Follow the Money investigation on the undetected dimension of fraud (European Chief Prosecutor).
• EDSPA economic figures: Chapters 9 and 9B of this version (ROI 164% over 5 years, net benefit +EUR 1,820M/year, cumulative break-even 2028, 9-State core).

12.1 Fundamental design principles

12.2 The organogram: a single institution with departments

Europe's experience with fragmented agency structures (the landscape reviewed in Chapters 6 and 11) recommends the opposite: a single institution with one direction, whose operational functions are organized as internal departments. This eliminates overlaps and makes the "agency against agency" game impossible. The judicial dimension is not an organ of the PP: it belongs to the EU judiciary (section 12.4).

12.3 The Council for Cognitive Protection

The Council is the political-strategic organ of the Protective Power. Its design resolves the tension between the technical impartiality required of a power that watches over manipulation and the political legitimacy without which the treaty reform will not be ratified. The solution is a mixed model with unequal voting:

Selection mechanism for the experts (merit + lottery)

1. Public call across the EU, with public eligibility criteria.
2. Evaluation by an independent committee following the European Research Council's standard.
3. Shortlist with at least three times the positions per Member State.
4. Public lottery audited before the Publications Office, notaries and media.
5. Ratification by the Plenary except on grounds of subsequent ineligibility.

Functions of the Council

• Legislative initiative in cognitive protection matters: the PP proposes; Parliament and the Council of the EU decide under the ordinary legislative procedure.
• Strategic direction: priorities, internal rules, multiannual plans.
• Supervision of the departments through the Executive Direction.
• Inter-institutional coordination with Executive, Legislature and Judiciary and with the Member States.

12.4 The judicial fit: specialized jurisdiction inside the judiciary

The PP will generate a significant volume of technical litigation — contested algorithmic audits, actions against its operational decisions, conflicts between platforms and cognitive rights. The temptation to solve it with a "Digital Protection Tribunal" owned by the PP was discarded in this version: a tribunal inside a power breaches the separation of powers and would confirm the perception of a power that is judge and party. The viable solution, backed by real precedents, has three pieces:

a) The administrative Board of Appeal (first guarantee, ACER model)

The PP's operational decisions (audit outcomes, designations of monitored platforms, information requests) will be challengeable before an internal but independent Board of Appeal, following the verified model of EU agencies' Boards of Appeal (ACER, CPDP): full review of the contested decision, with technical experts, appealable to the General Court. CJEU case law (Aquind v ACER, C-46/21 P) validates these boards as an appropriate rights-protection mechanism for complex technical decisions.

b) The Specialised Chamber in Digital Protection (second guarantee, inside the judiciary)

The volume and technicality of the litigation justify specialization, not a new tribunal. The proposal: a Specialised Chamber in Digital Protection of the General Court, created by ordinary regulation (no treaty reform needed), with:

c) The sanction: always external, always judicially reviewable

The PP has no sanctioning ceilings of its own. The real ceilings already exist in the acquis: up to 6% of worldwide turnover (DSA) and up to 7% or EUR 35M (AI Act). What is missing today is not higher ceilings: it is systematic detection and effective litigation. The PP's contribution is exactly that:

• Referral of evidentiary case files to the European Commission and the competent national authorities, at judicial standard of proof from day one.
• Litigation as a party before the Specialised Chamber and, via preliminary reference, before national courts.
• Publicity: documented manipulation is published — the reputational and market cost is, in most cases, the most immediate sanction.

12.5 The operational departments

12.6 The National Offices

12.7 The integration of the EDSPA (2033-2035)

12.8 Comparison: the agency versus the constitutional power

12.9 Conclusion: the architecture of a power for the 21st century

1. The Protective Power will be a single, protecting institution, not a sanctioning one: mixed Council with technical vote, single executive direction, four operational departments and national reach. Its weapon is light: evidentiary detection, continuous auditing and publicity.
2. Sanctioning stays entirely outside the PP: the existing ceilings (DSA 6% / AI Act 7%) are applied by the Commission and the national authorities; the PP refers case files and litigates as a party. No tribunal of its own, no sanctions committee, no fine money.
3. The judicial volume is solved inside the judiciary: administrative Board of Appeal (ACER model), Specialised Chamber of the General Court with dually qualified judges (UPC model) and an escalation clause conditioned on real demand (lesson of the Civil Service Tribunal).
4. The EDSPA does not disappear: it is integrated as the Department of Cognitive Protection, inheriting staff (1,500-2,000 employees), tools, standards and budgetary base (EUR 1,000M/year), leaving not a single day without protection between 2033 and 2035. And from there the endowment trajectory of a power begins (Ch. 15): ~3,000 staff in the core 2033-2035 → ~7,000-8,000 full ~2037 → more than 15,000 at maturity 2040+.

12.10 Verified sources and references

• Art. 19 TEU (judicial system), Art. 257 TFEU (specialised courts) and Art. 267 TFEU (preliminary rulings): EUR-Lex, consolidated texts.
• Specialised court precedent: EU Civil Service Tribunal (2004-2016), abolished by Reg. (EU, Euratom) 2016/1192 and reintegrated into the General Court.
• General Court expansion: doubling from 28 to 56 judges (2016-2019) to clear the backlog; CJEU 2024 statistics (1,706 new cases, 1,785 completed, 2,508 pending).
• Agency Boards of Appeal: ACER/CPDP model; CJEU, Aquind v ACER, C-46/21 P (full review of complex technical decisions).
• Dually qualified judges: Unified Patent Court (34 legally qualified + 51 technically qualified judges, 2022).
• Acquis sanction ceilings: DSA (Reg. 2022/2065, up to 6% of worldwide turnover) and AI Act (Reg. 2024/1689, up to 7% or EUR 35M).
• EU Digital Decade: target of 80% of adults with basic digital skills by 2030 (Decision 2022/2481); verified coverage ~55.6% (European Commission, JRC).
• European AI ecosystem: Mistral AI, Aleph Alpha, Apertus; InvestAI fund (EUR 200,000M) and EuroHPC AI factories.
• Economic framework: Chapter 17 (EUR 2,600M base / EUR 3,400M full) and own resources of Chapter 16.

13.1 The digital power vacuum: who controls the cognitive sphere today

The power vacuum is sometimes described with striking but unverifiable figures (a supposed "85% of Europeans receive news via algorithms", "70% of data outside EU jurisdiction", "90% of AI models American"). The thesis stands — effective control of the cognitive sphere lies with actors who answer to no citizens — but this book documents it only with what real sources support:

• Exposure to disinformation: the Eurobarometer of February 2025 found that 36% of Europeans report being exposed "often" or "very often" to content they consider distorted or false. Not a majority — but more than a third of the population exposed routinely.
• Erosion of informational trust: the Reuters Institute Digital News Report (2026 edition) documents a continued decline of trust in news, the rise of individual creators as a source, and the growth of news consumption mediated by AI assistants — an ecosystem where distribution is increasingly decided by opaque systems.
• Technological dependence: the Stanford AI Index (2026) confirms that the United States concentrate the production of top-tier AI models and China leads in scientific publications, while Europe advances with its own players (Mistral) but from a position of dependence. There is no verifiable "90%", but the structural asymmetry is real and measured.
• Manipulation of mixed origin: studies by the European Digital Media Observatory (EDMO) and the third FIMI report of the European External Action Service show that information manipulation campaigns in Europe combine domestic and foreign actors — there is no verifiable "60% of foreign origin", and the threat is not only geopolitical: it is also endogenous.

13.2 The framework of competences: five verbs and their limits

The competences are not organised as thematic blocks with sanctioning powers mixed in ("fines up to 10%", "bot neutralisation", "removal of fake accounts" have no place here). Everything is organised under the principle of pure protection: the same areas of action — cognitive protection, digital sovereignty, democratic protection, platform regulation, education — but described as protection functions with explicit limits.

13.3 The cognitive forensic scanner: the Protective Power's AI

One example must be rejected from the outset: "a European algorithm, audited and transparent, will decide which information is relevant". A State with an algorithm that decides informational relevance is precisely the computational censorship system this book denounces — merely with a different owner. This book replaces it with a central piece of the design:

This distinction is what separates the Protective Power from any ministry of truth. A power that filters is a censor with better branding. A power that detects, documents and publishes is a builder of public evidence: it changes the debate not because it controls what is seen, but because no one can manipulate in the dark without leaving a record.

13.4 The Citizen Checker: non-binding verification as digital education

Among the book's "citizen tools" there is a valuable intuition: giving citizens direct access to verification capacity. Developed carelessly, it would be dangerous: an "official verifier" issuing verdicts of truth would be a state oracle. Refined into its consensual form, it becomes a piece coherent with the pure protector. It is named the Citizen Checker and defines it with normative precision:

• Nature: a public verification service, non-binding and with no judicial value. The citizen pastes content (link, text, image, video) and receives a private report of indicators: manipulation techniques detected, forensic signals (image synthesis, amplification coordination), verifiable context. The report never issues "true" or "false" labels: the conclusion always belongs to the citizen.
• Permanent disclaimer: every report carries the warning that it is not a verdict, has no evidentiary value and does not replace one's own judgement. The Checker is a literacy tool in tool format, not a judge with a web interface.
• Privacy by design: access authenticated via eIDAS, but queries pseudonymised with no reversible link to identity; minimal retention under EDPS supervision; queries are never communicated to third parties, do not generate files and do not feed referrals. What a citizen checks is irrelevant to the PP: only that they learn to check matters.
• Anti-weapon symmetry: the forensic scanner is the same for all content — including that of the Protective Power itself, official communiques and electoral campaigns. A tool that only verifies others is a weapon; one that also verifies power is a mirror.
• Recognised risks: the Checker can be captured by media narratives ("the report says...") or generate over-trust replacing personal judgement. The mitigations are the indicator format (never verdicts), the permanent disclaimer and, above all, literacy: the service's goal is to become unnecessary.

13.5 The real regulatory context: why the acquis needs an enforcer

The urgency is not dated in hypothetical scenarios ("2029 European elections", "Brexit 2.0"). It is anchored in the real regulatory calendar, verifiable in the Official Journal of the EU:

The table's reading is the chapter's thesis: Europe has already written most of the law the Protective Power would enforce. The DSA orders transparency; the AI Act orders system oversight; the Data Act orders openness; the Shield defends elections. What does not exist is a public institution with the permanent mandate, technical resources and constitutional independence to enforce all of it systematically — to look inside the systems, document, publish and refer, every day, not only when a scandal erupts. That absent enforcer is exactly the set of competences described in 13.2.

13.6 The power transfer matrix: 2026 → 2035+

13.7 The constitutional limits: what it can and will never do

In a design whose central threat is manipulation, a public power with powerful forensic tools needs sharper limits than its competences. The chapter raises them to the rank of institutional definition — as binding as the five functions:

This radar is the institutional answer to the classic objection: "who watches the watchman?". The design's answer is threefold: (1) constitutional limits — the prohibitions of Figure 13.3 are inscribed in the power's own definition, amendable only through the Chapter 15 procedure; (2) symmetry — the scanner and the Checker apply to the PP itself; (3) specialised jurisdiction — every act of the PP is challengeable before the Specialised Chamber in Digital Protection and, on cassation, before the CJEU (Ch. 12). A power that cannot touch content, that audits itself with the same tools, and whose acts are judged by independent judges, has the answer built in.

13.8 Why these competences are urgent

Urgency does not depend on speculative scenarios. It depends on the verified convergence of three trends:

1. The regulatory window closes in August 2026. The AI Act reaches full applicability on 2 August 2026; the Data Act has been applicable since September 2025; the Art. 40 DSA delegated act is adopted. If no systematic enforcer exists during the acquis' years of force, the experience of European digital policy says the law stays on paper and platforms learn to meet the auditable minimum.
2. Manipulation is industrialising. The deepfakes and FIMI operations documented by the EEAS are increasingly cheap and effective; the third FIMI report shows coordinated campaigns against European democratic processes that current detection finds after the electoral damage is done.
3. Informational trust is structurally eroding. The Reuters Institute has documented for years the decline of trust and the shift towards non-editorial sources (creators, AI assistants) without verification standards. Every electoral cycle without cognitive protection infrastructure amplifies the spiral: less trust, more manipulability, less participation.

13.9 Conclusion: reclaiming digital power without conquering it

1. Digital power exists and is not in democratic hands. It is not controlled by a conspiracy: it is controlled by an opaque distribution infrastructure, AI models concentrated outside Europe, and actors — domestic and foreign — learning to manipulate it at low cost.
2. The acquis for reclaiming oversight is already written. DSA, AI Act, Data Act, European Democracy Shield, Tech Sovereignty Package: what is missing is a permanent, technical, independent enforcer. That is exactly the slot of the PP's competences.
3. The competences are five protective functions with constitutional limits. Detect, audit, publish, refer, educate — and never sanction, filter, prioritise or label. The cognitive forensic scanner generates evidence under dual human verification; platform algorithms keep deciding distribution, now in plain sight.
4. The Citizen Checker is a prudent provision. Non-binding indicator verification, pseudonymised, symmetric, in three phases and with no budget committed until an independent assessment says it is worth it.
5. The urgency admits no delay. The AI Act's full applicability (August 2026), the industrialisation of manipulation and the structural erosion of trust converge on the same years. Every year without an enforcer is an unprotected electoral cycle.

13.10 Verified sources and references

• Eurobarometer on the perception of disinformation (February 2025): 36% of Europeans report being exposed "often" or "very often" to distorted or false content.
• Reuters Institute Digital News Report (2026 edition): continued decline of trust and interest in news, rise of individual creators and of news consumption mediated by AI assistants.
• Stanford AI Index (2026): concentration of top-tier AI models in the US, Chinese leadership in publications, European advance (Mistral) from a dependent position.
• EDMO (European Digital Media Observatory): studies on information manipulation actors, combination of domestic and foreign actors in Europe.
• EEAS, third FIMI report: foreign information manipulation and interference campaigns against European democratic processes.
• DSA (Reg. 2022/2065): 6% ceilings, audits, and the Art. 40 delegated act on researcher access to platform data (adopted July 2025).
• AI Act (Reg. 2024/1689): ceilings of 7% of worldwide turnover or EUR 35M, full applicability on 2 August 2026.
• Data Act (Reg. 2023/2854): applicable since 12 September 2025.
• European Democracy Shield (European Commission, 12 November 2025) and Recommendation (EU) 2023/2829 on inclusive and resilient electoral processes (2023).
• Tech Sovereignty and Cloud AI Package (European Commission, 3 June 2026): Cloud and AI Development Act, Cloud Sovereignty Framework, dependency reduction, European infrastructure.
• Shoshana Zuboff (2019), The Age of Surveillance Capitalism and work by the Oxford Internet Institute: analytical framework of the power of informational infrastructure.
• Institutional design: Chapters 8 (human-AI dual verification), 12 (architecture, specialised jurisdiction, departments), 15 (legal basis) and 16 (own resources).

14.1 The real precedents: Europe has already done this three times

The chapter grounds its credibility not in documents that do not exist ("White Paper on Constitutional Reform 2024", "Report on the Feasibility of a 4th Power 2023", "Reflections on the Future of the EU 2022"), but in a verifiable precedent of European law and institutional history for every piece of the process:

14.2 The master timeline: five phases, 2027-2035

Phase 1: EDSPA operational (2027-2030)

The nine founding states launch the agency with the base budget of EUR 1,100M/year. The goal is not only to protect from day one, but to generate the evidence that makes reform inevitable: audits published, files referred, elections protected (including the 2029 European elections), and the 164% ROI and EUR 1,820M/year net benefit documented with the Chapter 9B methodology. The EDSPA does not ask for faith: it produces auditable results.

Phase 2: Assessment and proposal (2030-2031)

A triennial EDSPA impact report with external verification; a Europe-wide public consultation with randomly selected citizen panels (the method already tested by the Conference on the Future of Europe); and a formal European Commission proposal for an enhanced cooperation institutionalising the Protective Power. The argument is Chapter 11's: the EDSPA works, but it can be revoked by a political majority — cognitive protection needs constitutional anchoring.

Phase 3: Legal route (2031-2033)

Council authorisation of the enhanced cooperation (qualified majority, after meeting the Art. 20 TEU requirements), negotiation of the constitutive convention among participating states, signature and ratification with a 24-month margin — the real time Lisbon took, including one national setback. For states whose constitutions require referendums, the Danish UPC precedent (2014) shows these processes can also be won at the ballot box.

Phase 4: Entry into force of the core (~2033)

The Protective Power is born with the states that have already ratified — no fewer than 9 —, with provisional application from signature so that the organs (Council of Cognitive Protection, executive directorate, start of the Specialised Chamber) are built and staffed on day one, exactly as the UPC hired its judges before opening its doors.

Phase 5: Gap-free integration (2033-2035) and gradual accession (2033 → horizon)

By a successor-organisation act modelled on Regulation 2016/794 (Europol), the EDSPA becomes the Department of Cognitive Protection of the PP: staff, seat, systems, files and legal bases transfer as a block, and the agency keeps all its competences until the last day. In parallel, the other three departments and the Chapter 12 specialised jurisdiction are built. The other states accede at their own pace, each with its own ratification — the Schengen model: it started with fewer states than the Community and today covers almost all of them, including non-EU countries.

14.3 The double constitutional route: why nine and not twenty-seven

The underlying reason is not tactical but democratic: demanding unanimity to start means the most reluctant — or most manipulated — state decides whether the others may protect themselves. Enhanced cooperation inverts that logic: those who want protection get it soon, and those who wait block no one. And experience teaches that what works ends up attracting: Schengen, the euro and the UPC all grew after being born small.

14.4 What is transferred — and what is left untouched

14.5 The continuity guarantee: not one day of gap

14.6 Obstacles and strategies

14.7 Success indicators of the transition

The chapter promises no invented quantitative results ("-40% disinformation by 2030", "70% citizen support", "12→20→27 countries"). It offers verifiable process indicators — things that can actually be counted and audited:

• Protection: algorithmic audits completed and published per year; files referred to the Commission and resulting sanctioning decisions; the 2029 and 2034 European elections covered without serious documented-and-unanswered interference incidents.
• Economic credibility: ROI of 164% and net benefit of EUR 1,820M/year externally verified (Ch. 9B methodology), budget executed with deviation below 5%.
• Expansion: states acceding to the enhanced cooperation (9 → 15 → 20+); population coverage of cognitive protection; accessions after each well-protected electoral cycle.
• Social capacity: progress of cognitive literacy towards the 90%+ target by 2038 (from the Digital Decade's real baseline: 80% target for 2030, ~55.6% verified today).

14.8 Conclusion: a transition designed not to fail

1. The process does not ask for a political miracle: it asks for nine determined governments. Art. 20 TEU allows starting with the founding core; Schengen, the euro and the UPC show that successful European integrations are born small and grow through results.
2. The timeline is anchored in real durations. 24 months of ratification (Lisbon), provisional application (UPC), gap-free successor organisation (Europol): every phase copies a precedent that already happened, not a hope.
3. The transition does not dilute the pure protector: it consolidates it. No sanctioning powers are transferred because they exist at neither level; fines stay with the Commission and the EU's coffers; the PP reaches 2035 with more coverage, more departments and the same weapon as always: light.
4. The all-27 route is not abandoned: it is rescheduled. Unanimity is the consolidation horizon at a future Convention — when existing and working is the best argument for a Treaty that cannot be approved today.

14.9 Verified sources and references

• Art. 48 TEU (ordinary and simplified revision procedures) and Art. 20 TEU (enhanced cooperation: minimum 9 states): EUR-Lex, consolidated texts.
• European Parliament resolution of 22 November 2023 triggering Art. 48 TEU for the first time and asking for a Convention to be convened (vote: 291 in favour, 274 against, 44 abstentions; the 9-Jun-2022 resolution asking for the same was adopted 355:154:48).
• Conference on the Future of Europe (2022): final report, 49 proposals with more than 300 measures, randomly selected citizen panels; proposal 39 on ending unanimity and 38 on European referendums.
• Europol: Regulation (EU) 2016/794, which turned the intergovernmental organisation into an EU agency on 1 July 2017, "replacing and repealing" its founding acts, with staff and activities transferred.
• Unified Patent Court: 2013 Agreement, provisional application from 2022, entry into force 1 June 2023 with 17-18 states; Danish accession referendum (25 May 2014, 62.5% in favour).
• Treaty of Lisbon: signed 13-Dec-2007, in force 1-Dec-2009; ratification completed with the Czech deposit of 13-Nov-2009; Irish referendum of 2008 (rejection) and 2009 (approval).
• Enhanced-cooperation precedents: Rome III divorce regulation (Reg. 1259/2010, 14→17 participating states) and the unitary patent.
• Schengen and the euro as precedents of gradual accession and variable-geometry integration.
• The book's own frameworks: Ch. 9B (ROI 164%, EUR 1,820M/year), Ch. 6 (founding core of 9), Ch. 11 (EDSPA as necessary step), Ch. 12 (architecture and 2033-2035 integration), Ch. 13 (pure protector competences), Ch. 16 (own resources) and Ch. 17 (EUR 2,600M/3,400M).

15.1 The hierarchy of legal bases: three steps towards the 4th power

Nothing in the PP's design requires inventing new law: it requires stepping up a hierarchy that already exists. Each step has its basis, its rank and its reversibility — and the book's argument is that only the third guarantees protection that does not depend on an annual political majority.

15.2 The principle of protection is already in the Treaty

The most common objection — "there is no legal basis for such a power" — confuses two questions: whether the EU can protect the cognitive sphere, and whether the EU must do so at constitutional rank. The first is already answered by existing law:

The case law that already protects the cognitive sphere

• CJEU, Digital Rights Ireland (C-293/12 and C-294/12, 2014): invalidated the Data Retention Directive because mass, indiscriminate retention interferes in a particularly serious manner with fundamental rights — the Court already said that total surveillance of digital life is incompatible with the Charter.
• CJEU, Google Spain (C-131/12, 2014): the right to erasure against the search engine recognises that the individual controls their informational footprint: informational autonomy is an operative right, not a theoretical one.
• CJEU, Schrems II (C-311/18, 2020): the Privacy Shield fell because transferring data to third countries without equivalent protection is unlawful — the sovereignty of personal data is enforceable.
• The common line: the person's cognitive sphere is legally protectable — the CJEU has defended it case by case. The PP proposes what that case law's logic already points to: systematic protection, not only reactive.

The financial basis: Art. 311 TFEU

Chapter 16's financial autonomy is grounded in Art. 311 TFEU: the Union provides itself with own resources through a Decision adopted unanimously and approved by the states under their constitutional rules — the same mechanism as Decision 2020/2053 that financed the post-pandemic recovery. A stable constitutional power needs exactly that: income it does not renegotiate with each annual budget. The PP is not financed with fines and taxes; financing is the matter of Chapter 16: own resources yes, fines as own income never — the revenue circle works indirectly (see 15.5).

15.3 Montesquieu and the 4th power: the objection and the answer

The answer has two parts. The historical one: Montesquieu did not enumerate a closed list in The Spirit of the Laws (1748) — he described the principle that power must check power, and constitutions have multiplied the organs that embody it: constitutional courts, independent central banks, ombudsmen, fiscal authorities. None existed in 1748; all are democratic orthodoxy today. The functional one: the three classic powers now operate in a cognitive sphere they do not control — and a power that does not control the terrain of the public opinion that legitimises it balances nothing: it is blind. The PP does not add a competitor to the three: it ensures the three can see. And unlike a sanctioning 4th power — which would be a real danger — this one is designed without the tool that would make any power fearsome: it cannot punish, only illuminate (Ch. 13). The separation of powers is not broken by a power that publishes; it would be broken by a power that silences, filters and punishes at once.

15.4 The endowment of a power: bringing democracy into the digital era

European democracies remain anchored in the analogue era. Their three powers were designed — and endowed — when public opinion was formed in the square, the press and television: the legislature to deliberate in chamber, the executive to administer territories, the judiciary to try cases. Today wills are formed in opaque recommendation systems, operated mostly from another jurisdiction, and none of the three powers watches that terrain. That is the constitutional anachronism of the 21st century, and correcting it is not creating "one more agency": it is endowing democracy with the power it lacks in order to exist in the digital era. And a power is endowed the way powers are endowed.

Offices are not handed out by decree with pre-assigned cities, nor with a magic figure of "10,000+ employees". The chapter keeps the three categories of presence — proximity, coordination, excellence — under the criterion of merit and open call, and replaces any magic figure with a staged trajectory in which every jump has an accountable trigger: accessions of states, coverage of the mission, volume of threat and the expansions of mandate the legislator authorises over time — as has happened to every European power.

Why the figure is that of a power, not of an agency

Three masses of staff make the trajectory inevitable, and none of them is bureaucracy:

• The real scale of the operation. Continuous forensic surveillance of the cognitive sphere of ~450 million people across 27 states and 24 official languages is an operation of shifts and continuous coverage: forensic scanner operators, human verifiers (Chapter 8's dual verification is deliberately non-automatable — it is the heart of the design), platform auditors by region and language, national offices that at maturity operate with 100-150 people. This block alone already exceeds the ceiling of the old "agency" projection.
• The cognitive education corps. The target of literacy of 90% or more by 2038 is a programme the size of an education system: trainers, instructional designers, coaches, coordination with the school systems of 27 countries. Education does not scale with software alone — it scales with teachers.
• The organic growth of every power. All European powers have grown in mandates and staff over time — the Commission from a few thousand to ~32,000; Frontex from dozens of analysts to a statutory corps of 10,000 set for 2027. The PP will not be the exception: between 2033 and 2040 the legislator will authorise expansions — Chapter 13's Citizen Checker is only the first one foreseen, and its staffing needs are already inside the maturity horizon, without a separate figure. What does not grow is the nature of its powers: Chapter 13's constitutional limits freeze what it can do; the stages only scale how much of the same.

15.5 The financing of a power: the virtuous revenue circle

And who pays for a power of more than 15,000 staff? The question has two answers — one accounting, one doctrinal — and both play in favour of the design.

The accounting one: the revenue circle. Every forensic file the PP refers and the Commission turns into a sanction (DSA up to 6%, AI Act up to 7%) goes into the EU's general coffers — and from those coffers, via the own resources of Art. 311 TFEU, the PP's budget is endowed. The circle is virtuous and structurally separated, as the pure protector demands: the power that detects never collects what it detects — but the Union that endows it does collect what its work produces. And the magnitude is no longer theory: the Commission has imposed fines of billions of euros under the digital acquis (Google Shopping, Android, Apple...), and each one sharpened the discipline of the European market. A systematic PP would multiply that flow: not as own income of the power (never), but as a flow that makes the whole Union gain from every file. The PP's activity, which costs, is the same activity that raises revenue.

The doctrinal one: a power does not need to be viable — it needs to fulfil its mission. Nobody asks whether the judiciary "is profitable", or whether parliament "pays for itself": they are endowed with what their constitutional function requires, because without them there is no democracy that can afford the luxury of calculating. The PP belongs to that category: it is the condition for democracies to exist in the digital era, and that is why it is endowed according to its mission, not according to its profitability. That in addition — as Chapter 9B demonstrated with verifiable methodology — it yields a 164% ROI and a net benefit of EUR 1,820M/year makes it an almost unique case in constitutional history: a power that justifies itself without economics and that, nevertheless, economics endorses. Viability is the cherry on the cake; the cake is protected democracy.

15.6 Conclusion: the legal basis exists; what is missing is rank — and endowment of power

1. The EU already holds the mandate to protect the cognitive sphere. Values (Art. 2 TEU), objectives (Art. 3 TEU), rights (Charter, Arts. 7, 8, 11, 47), competences (Arts. 16 and 114 TFEU, Art. 352 if needed) and case law (Digital Rights Ireland, Google Spain, Schrems II): cognitive protection is the pending enforcement of existing law.
2. What is missing is not basis: it is rank and an enforcer. For that there are Art. 48 TEU (the full route) and Art. 20 TEU (the founding route) — Chapter 14's double route.
3. Montesquieu's objection is answered by the design. The principle is that power check power; the PP adds the missing counterweight and arrives without the tool of danger: it does not sanction, it does not filter, it only illuminates, with independent jurisdiction behind every act.
4. And it is endowed as what it is: a power. Democracies anchored in the analogue era need the power that brings them into the digital era — with an endowment trajectory that goes from the 1,500-2,000 inherited from the EDSPA to more than 15,000 staff at maturity 2040+, financed by the revenue circle of its own files and by own resources, and turned into Europe's largest employment pole of non-automatable human judgement against AI. A power does not need to be viable: it needs to do its job well — and this one, besides fulfilling its mission, pays for itself.