Por: Carlos A. FERREYROS SOTO

Doctor en Derecho

Universidad de Montpellier I Francia.

cferreyros@hotmail.com

RESUMEN

La gobernanza mundial de la IA en 2025 se desarrolla en un contexto geopolítico inestable, marcado por tentativas de coordinación internacional que revelan importantes discordias entre las potencias mundiales . La ausencia de un cuadro global coherente conduce a una proliferación de iniciativas, foros y coaliciones con prioridades divergentes. La tensión entre unilateralismo y multilateralismo fragiliza la gobernanza, permitiendo que surjan intereses privados poderosos con una influencia desproporcionada sobre las normas y el acceso a las infraestructuras críticas .

Además, las brechas entre el desarrollo tecnológico y la regulación son cada vez más amplias. La Unión Europea, pionera tradicional en la regulación, tiene ahora una influencia cada vez menor en este debate. Para seguir siendo un actor global creíble, Europa debería replantear su estrategia, al mismo tiempo que abordar las deficiencias internas en la aplicación de la normativa y la disminución de su influencia externa.

El presente artículo examina también la dinámica actual de la gobernanza y argumenta que la ambición regulatoria por sí sola ya no es suficiente. Para mantener su credibilidad, la UE asociar la regulación normativa con las competencias de investigación y desarrollo y establecer alianzas y coaliciones basadas en intereses estratégicos compartidos, en lugar de políticas y regulación inciertas.

Se adjunta el texto en inglés. La cita oficial al documento además del enlace al texto original es la siguiente: Instituto Universitario Europeo y Cantero Gamito, M., AI Governance and the EU’s Strategic Role in 2025 , Instituto Universitario Europeo, 2025, https://data.europa.eu/doi/10.2870/2955242

A fin de acceder a normas similares y estándares europeos, las empresas, organizaciones públicas y privadas positivas en asesorías, consultorías, capacitaciones, estudios, evaluaciones, auditorías sobre el tema, sírvanse comunicar al correo electrónico: cferreyros@hotmail.com

___________________________________________________

AI GOVERNANCE AND THE EU’S
STRATEGIC ROLE IN 2025

Author:

Marta Cantero Gamito

Florence School of Transnational Governance

ISSUE 2025/13
AUGUST 2025

EXECUTIVE SUMMARY

The current global conversation on AI governance is taking place

within an intense and shifting geopolitical setting. As such, the

ongoing attempts to coordinate governance through summits

and other international initiatives are revealing important

disagreements among world powers. The EU, once a regulatory

leader, faces weakening influence in this conversation. To remain

a credible global actor, Europe should rethink its strategy while

dealing with internal enforcement gaps and declining external

leverage. This policy brief examines current governance dynamics

and argues that regulatory ambition alone is no longer sufficient.

To remain credible, the EU needs to connect rulemaking with

industrial capacity and build coalitions shaped by shared strategic

interests rather than rhetorical alignment.

Author:

Marta Cantero Gamito | Research Fellow, Florence School of Transnational Governance, EUI

Views expressed in this publication reflect the opinion of individual authors and not those of the European

University Institute.

1. CONTEXT AND STAKES¹

In 2025, aspirations for global coherence in AI

governance seem unfeasible. Recent summits

such as the Paris AI Action Summit and the

Munich Security Conference, rather than

building consensus, have exposed significant

disagreements. At the same time, the gap

between AI development and regulation is

widening. Also, since the EU proposed the

Artificial Intelligence Act (the so-called ‘AI

Act’) in 2021, the geopolitical climate has

dramatically changed. The proposal came in a

world still buffered by the regulatory influence

of the GDPR. However, AI is more than data.

It involves software, hardware, semiconductors

and, more broadly, network infrastructure and

security architectures. As such, AI is a deeply

political technology and a strategic governance

tool. Therefore, tensions surrounding its

regulation and governance are no longer

about compliance but also a competition over

control and who gets to write the rules.

Now, however, the illusion of control is fading.

The proliferation of AI governance venues has

created overlapping mandates and fragmented

commitments. Instead of convergence, there

has been an increase in competing authorities

and institutional complexity.² Yet, the risk of

ineffective AI governance is not just duplication

or inefficiency. The concern is rather the

systematic marginalisation of those unable to

follow, let alone shape, these initiatives. Often,

states and institutions without the resources

to engage across multiple agendas are left

behind. This is particularly true for Global

South countries whose voices are sidelined by

capacity asymmetries.³

Meanwhile, the boundary between governance

and geopolitical strategy is increasingly blurred.

What was once about market harmonisation

or data protection now concerns power and

sovereignty. Although the EU continues to

position itself as a value-based rulemaker, it

struggles to project that identity in a world of

largely influenced by the United States and

China.⁴ The digital ‘third way’ Europe once

offered is under pressure, if not in retreat.⁵

In this shifting order, questions about who

sets the rules, who enforces them, and

who gets left behind are no longer abstract

governance questions but critical geopolitical

challenges. Europe’s role in this context is at

once decisive and precarious. The EU aspires

to act as a bridge by upholding democratic

accountability while engaging strategically

with global powers. However, its success

depends on staying normatively coherent

and geopolitically relevant. This brief explores

the resulting landscape of convergence and

divergence in AI governance and Europe’s

place in it.

2. FLOCKING BEHAVIOUR IN AI

GOVERNANCE

Much of the current realignment in AI

governance is not entirely intentional. In fact,

when institutional responses do emerge, they

are often reactive, imitative, and driven more

by perceived necessity than by deliberate

design. Traditional accounts of international

cooperation assume a degree of strategic

coordination and institutional rationality.⁶

However, in the AI domain, what we observe

increasingly resembles flocking, not structured

coordination.

Borrowed from behavioural ecology, flocking

describes the tendency of animals (mainly birds)

to follow perceived leaders, often without fully

evaluating the implications or alternatives. A

5% shift in a certain direction by a dominant

actor direction can pull the entire group into

a new trajectory. The same is observed in AI/

digital governance, where early movers create

gravitational pulls that shape the regulatory

landscape beyond their borders.

This imitation reproduces and amplifies

asymmetries of power. Those who lead

also define the terms of engagement. For

many states, especially in the Global South,

alignment with dominant models is less about

shared values and more about infrastructural

dependency, capacity gaps or geopolitical

pressure (or all of the above). In such cases,

alignment can serve as a proxy for legitimacy,

even when the underlying distribution of

agency remains deeply uneven.

Calls for regulatory interoperability, particularly

in global trade, now collide with an increasingly

fragmented global landscape. The move away

from hierarchical rulemaking toward a more

dispersed legal order has increased complexity

without resolving questions of authority.

While this shift might signal a higher-intensity

pluralism, it also strains the coherence required

for effective global coordination. In this context,

technocratic governance has emerged not just

as a Western trend, but as a structural response

to the perceived ungovernability of deeply

interdependent but politically divided systems.⁷

Whether or not this is a global phenomenon

remains an open question, but its implications

are evident in the widening gap between

formal participation and real influence.

Flocking also masks the role of ‘predators’ which

are actors that can reshape the trajectories of

others through leverage rather than persuasion.

For AI governance, these may be powerful firms

controlling access to large-scale computational

infrastructure, foundational models, or

standard-setting processes.⁸ Their ability to

frame risk, define technical parameters, or

even set governance priorities often bypasses

formal institutional checks.

Behavioural science sheds further light on the

origins of reactive behaviour. Under uncertainty,

actors tend to conform, especially when losses

(e.g. in competitiveness or security) loom

larger than potential gains.⁹ In the context of

AI, risk narratives exacerbate these dynamics

by creating perceived urgency and limiting

the space for deliberation.¹⁰ As a result, the

global regulatory landscape often resembles a

cascading response system in which imitation

replaces deliberate policy direction. The result

is a governance ecosystem driven less by

principled coordination than by reputational

pressure.

What this process generates is the appearance

of supranational alignment. Superficial signs

of cooperation, such as countries signing

joint statements, adopting similar language,

or participating in the same forums, do not

guarantee meaningful agreement or shared

commitment. This points to the risk of

performative governance, where coordination

is claimed without enforceable obligations or

inclusive participation. Therefore, to assess

where digital governance is headed, we need

to examine the architectures that underpin it.

3. GLOBAL ARCHITECTURES IN

DIGITAL GOVERNANCE: WHO HAS

GRAVITAS?

What began as a race to regulate has become

a competition to shape and control markets

and critical infrastructures. In the absence

of a coherent framework for AI governance,

a fragmented space of initiatives, summits,

institutes, frameworks, and coalitions has

emerged, with each forum advancing its own

priorities, definitions, and claims to legitimacy.

However, while pluralism is often seen as a

strength, it has led to architectural incoherence.

Rules proliferate without coordination,

responsibilities are diffused, and important

gaps remain unaddressed. Forums like Internet

Governance Forum (IGF), once celebrated as

an unparalleled multistakeholder format, have

been sidelined by more agile and exclusive

coalitions, which accelerate coordination

among powerful players but often do so at the

expense of transparency and inclusion.

At the centre of this fragmentation lies a

deeper tension between multilateralism and

multistakeholderism. The fate of the IGF’s

mandate since the adoption of the Global Digital

Compact in September 2024 is symptomatic

of this conflict. While the former centres state

legitimacy, the latter distributes influence

among diverse stakeholders. In theory, this is a

pragmatic solution to the complexity of digital

power. In practice, though, it can complicate

(and even obscure) structures and questions

of accountability or even open the door to

governance capture, as seen in other sectors.¹¹

This tension is especially visible in AI governance,

where control over infrastructure and agenda-

setting power is heavily concentrated in

the hands of a few private frontier labs and

hyperscale cloud providers. These actors are

invited into governance spaces as technical

experts, but their dual role as both rulemakers

and market actors creates conflicts of interest.

In this sense, as Kate Crawford has noted, “AI is

in its empire era,” characterised by expansionist

ambitions, massive investment in extractive

infrastructure, and limited accountability.¹²

As a result, far from convergence, AI

governance models find themselves in conflict,

driven by competing agendas. Some prioritise

coordination, others fragmentation. Some

emphasise precaution, others acceleration.

Some claim democratic accountability,

others strategic utility. The result is a layered

geopolitical landscape marked by parallel

architectures built on incompatible assumptions

about sovereignty, risk, and/or legitimacy.

The Paris AI Action Summit highlighted

this fragmentation. While it invoked

multistakeholderism and global coordination,

the absence of key countries (including the US

and the UK) from its final declaration exposed

the fragility of any substantive consensus.

Also, the summit’s rhetoric of inclusion stood

in contrast with the reality. Voices from the

Global South were present (for instance, on

panels), but their influence on decision-making

remained peripheral.

Besides,   this   divergence   is   not   merely

procedural. Scholars like Sean O hEigeartaigh

describe the current moment as a ‘pre-AGI

diplomatic phase,’ where informal coordination

among dominant actors replaces formal

institutional rules. In this context, informal elite

consensus between labs, leading states, and

technical experts may shape the future more

decisively than treaty-based multilateralism.¹³

The question is, then, not only who sits at the

table, but whose agenda gets embedded in

governance frameworks. In a 2021 paper, Dafoe

and Carlier called for a “constitutional moment”

in an AI governance landscape shaped by

high uncertainty and long-term normative

consequences.¹⁴ Yet, the core concern persists:

can such moments deliver institutions that

remain epistemically independent?

In response, some attempts to mediate these

tensions are emerging. One is a proposal

for a jurisdictional certification model for

AI governance, in which states voluntarily

recognise and interoperate with one another’s

domestic AI regulations based on shared

principles.¹⁵ This approach avoids the

rigidity of a centralised treaty model while

enabling coordination through regulatory

interoperability. It mirrors the logic of mutual

recognition in global trade but is adapted

to a risk-sensitive and values-diverse digital

domain. Its feasibility, however, depends on

trust and transparency, both currently in short

supply.

Ultimately, the debate is not just over who

governs but how. The deeper tension is

between the ideal of interoperable and

inclusive governance and the political reality

of fragmentation and strategic competition.

This is highly visible in Europe, where the

EU’s ambition to act as a normative bridge is

increasingly constrained by weakened leverage

and growing uncertainty about Europe’s

position in the current global order.

4. EUROPE’S STRATEGIC ROLE

Can the EU still shape digital governance, or

is it being sidelined? In a world that is growing

less inclined to flock around its regulatory

model, the EU’s status as a norm entrepreneur

is challenged by internal fragmentation and

the waning traction of the ‘Brussels Effect.’

The strategic position that once allowed the

EU to mediate between divergent approaches

(e.g. those of the US and China) is increasingly

difficult to maintain. In the wake of the Paris

AI Action Summit, the EU finds itself gradually

growing more isolated. The refusal of the US

and the UK to endorse the final declaration,

as well as the muted response from frontier AI

labs, exposed the limits of Europe’s influence.

At the same time, the EU is facing increasing

enforcement dilemmas at home. While the

AI Act, the Digital Services Act (DSA), and

the Digital Markets Act (DMA) are recent

achievements in global digital regulation,

challenges related to their implementation

reveal a growing mismatch between European

rulemaking ambition and enforcement capacity

(and, more critically, political will). At the 2025

Munich Security Conference, US Vice President

JD Vance openly criticised Europe’s “excessive

regulation” and hinted at retaliatory measures if

the EU continued to target American platforms.

Growing lobbying efforts and the political cost

of enforcement have all contributed to the view

of these regulations as ‘rules without teeth.’

The quiet shelving of the AI Liability Directive

earlier this year also signals regulatory fatigue,

both administrative and geopolitical.

The EU’s soft power is also under strain. The

EuroStack initiative, an industry-proposed

sovereign digital infrastructure alternative to

Silicon Valley or Beijing, has not yet materialised.

Although the EU has committed to long-term

investment in AI research, compute capacity,

and cloud sovereignty, it still lacks globally

competitive firms to project its regulatory norms

through market reach. Europe has limited

access to foundational models and continues

to rely on external compute infrastructures.

These structural limitations undermine the EU’s

ability to translate regulatory ambition into

global influence.

The AI Continent Action Plan, released in April

2025, aims to address this issue. It reflects a

shift from purely rules-based governance

toward a more active industrial strategy.¹⁶

Building on strategic autonomy (the EU’s

preferred euphemism for digital sovereignty),

the plan seeks to coordinate investment and

develop interoperable infrastructure in an

attempt to strengthen Europe’s position across

the AI value chain. Politically, it signals a more

pragmatic turn, recognising that normative

influence requires industrial leverage. Its

success, however, will depend on aligning

member state priorities and attracting private

sector engagement.

The challenge is also structural. Earlier EU

successes in digital governance were achieved

in a more stable technological and geopolitical

context. Regulatory action is now driven

by urgency and a perceived need to assert

sovereignty in a race that is already underway,

rather than by the demand to solve pressing

societal problems and protect rights. This is

visible in the AI Act’s language, which mirrors

the rhetoric of risk and strategic autonomy.

Given these dynamics, the EU faces a critical

decision. Should it double down on its goal

to lead global digital governance, or should

it focus instead on building industrial capacity

and agreeing on a common EU digital policy

for its citizens?

Some call for realism and the acknowledgement

that Europe must first consolidate its internal

capacity before projecting global leadership.¹⁷

Others warn that retreating from global

engagement would accelerate the EU’s

marginalisation and leave the normative ground

to authoritarian models.¹⁸ In either case, the

EU’s position will ultimately depend less on

its regulatory volume and more on its political

credibility and technological relevance.

5. POLICY RECOMMENDATIONS IN

AN IMPOSSIBILITY TRILEMMA

AI governance is currently constrained by an

impossibility trilemma. In other words, no policy

can simultaneously prioritise state sovereignty,

individual autonomy, and unrestricted

innovation. Advancing two of these principles

typically undermines the third. Protecting

sovereignty and individual autonomy often

restricts innovation, promoting innovation and

individual autonomy weakens state control, and

the pursuit of state control and technological

leadership often leaves little room for dissent,

privacy, or individual agency.¹⁹

The EU is at the core of this tension because

it currently seeks to simultaneously regulate

innovation in the public interest, safeguard

individual rights, and retain strategic autonomy.

To complicate things further, it must do so in a

global environment shaped by infrastructures

that it cannot fully access and companies

that it does not control. In this context, policy

recommendations should move beyond

chasing impossible balances and instead focus

on asserting a clear normative stance that

projects credibility and direction amid shifting

geopolitical and technological constraints.

1. Enforcement must be treated as

existential. The EU’s influence on the

world stage has thus far largely relied on

enforcement, but this credibility is eroding.

Geopolitical credibility requires sustained

investment in enforcement capacity, which

entails stronger national authorities with

the budgets and technical expertise to act

effectively, sound cross-border enforcement

mechanisms across member states, and

active support for civil society litigation in

the public interest.

2. Public investment must reflect public

values. European AI innovation cannot

depend on non-EU entities whose business

models run counter to European interests.

EU funding schemes should be deliberately

aligned with the Union’s normative

standing. This includes conditioning public

funds on model openness and democratic

oversight. Moreover, the EU must promote

investment in accessible compute capacity

and sovereign AI infrastructure. Here,

instruments like the European Chips Act,

the Cyber Resilience Act, and the Data Act

can provide critical regulatory support to

ensure that Europe retains control over its

hardware, cybersecurity, and data flows.

3. Incentives-driven coalitions should
   be  prioritised  over  extraterritoriality.

While universal convergence now seems

unlikely, value-aligned coalitions remain

possible. The EU should lead in creating

interoperable, flexible arrangements that

reflect its core values, such as jurisdictional

certification models consisting of voluntary

mutual recognition agreements among

states whose domestic AI regimes meet

shared standards. This approach mirrors

mutual recognition in trade law but adapts

it to a value-based governance framework.

It would allow Europe to avoid a potential

backlash against extraterritorial imposition,

shaping global norms through alignment

rather than dominance.

4. Governance infrastructure must support

epistemic independence. While the

European AI Office plays a central role in

implementing the AI Act and conducting

technical evaluations, democratic

resilience requires distributed expertise,

as seen, for instance, in the field of climate

governance with the Intergovernmental

Panel on Climate Change (IPCC). The EU

could support the creation of a European

network for AI foresight and risk: a publicly

funded but institutionally independent

consortium of academics, civil society, and

technical experts. This network would offer

critical foresight, signal slow-moving risks,

and conduct open evaluations of high-

impact models and infrastructures while

complementing (rather than replicating) the

AI Office.

5. Governance must be anticipatory, not

reactive. This requires embedding digital

governance into Europe’s foreign, industrial,

and security strategies. In this regard, the

creation of a Digital Geopolitics Council as

a permanent advisory body for emerging

tech foresight would help the EU reconcile

regulatory frameworks with its global

positioning and industrial policy.

6. CONCLUSION

Flocking behaviour in digital governance

highlights both the unstable nature of global

coordination and the systemic pressures that

shape it. Understanding underlying behavioural

patterns is needed to shape forward-looking

governance rather than simply reacting to it.

Yet, if AI governance continues to be treated as

a mere regulatory tool rather than as a strategic

end, current frameworks may fail to capture the

shifting balance of power.

In light of this, the EU’s reliance on regulatory

projection is reaching its limits as a tool of

influence. Europe cannot afford to continue

externalising its industrial dependencies if it

expects to retain regulatory power. Instead,

to remain a credible actor in global digital

governance, the EU needs to support its

regulatory ambition with industrial capacity

while forging coalitions that help to hold the

flock together.

The School of Transnational Governance (STG) delivers teaching and

high-level training in the methods, knowledge, skills and practice of

governance beyond the State. Based within the European University

Institute (EUI) in Florence, the School brings the worlds of academia

and policy-making together in an effort to navigate a context, both

inside and outside Europe, where policy-making increasingly transcends

national borders.

The School offers Executive Training Seminars for experienced

professionals and a Policy Leaders Fellowship for early- and mid-

career innovators. The School also hosts expert Policy Dialogues and

distinguished lectures from transnational leaders (to include the STG’s

Leaders Beyond the State series which recorded the experiences of

former European Institution presidents, and the Giorgio La Pira Lecture

series which focuses on building bridges between Africa and Europe). In

September 2020, the School launched its Master-of-Arts in Transnational

Governance (MTnG), which will educate and train a new breed of policy

leader able to navigate the unprecedented issues our world will face

during the next decade and beyond.

The STG Policy Papers Collection aims to further the EUI School of

Transnational Governance’s goal in creating a bridge between academia

and policy and provide actionable knowledge for policy-making. The

collection includes Policy Points (providing information at-a-glance),

Policy Briefs (concise summaries of issues and recommended policy

options), and Policy Analyses (in-depth analysis of particular issues).

The contributions provide topical and policy-oriented perspectives on

a diverse range of issues relevant to transnational governance. They

are authored by STG staff and guest authors invited to contribute on

particular topics.

______________

1 Marta Cantero Gamito is a Research Fellow at the EUI’s Florence School of Transnational Governance and Professor of Information Technology Law at the School of Law of the University of Tartu (Estonia).

2 Nye, J. S. (2014). The regime complex for managing global cyber activities. Global Commission on Internet Governance Paper Series: No. 1, May 2014. On the description of a broader phenomenon of institutional complexity in global governance, see Abbott, K. W., & Faude, B. (2022). Hybrid institutional complexes in global governance. The Review of International Organizations, 17(2), 263-291.

3 Heeks, R. (2022). Digital inequality beyond the digital divide: conceptualizing adverse digital incorporation in the global South. Information Technology for Development, 28(4), 688-704.

4 Csernatoni, R. (2024). Charting the Geopolitics and European Governance of Artificial Intelligence. Carnegie Endowment for International Peace. Accessible at https://carnegieendowment.org/research/2024/03/charting-the-geopolitics-and-european-governance-of-artificial-intelligence?lang=en.

5 Bradford, A., Kelemen, R. D., & Pavone, T. (2024). Europe Could Lose What Makes It Great. Foreign Affairs (April 25, 2025). Accesible at https://www.foreignaffairs. com/europe/europe-could-lose-what-makes-it-great

6 Koremenos, B., Lipson, C., & Snidal, D. (2001). The rational design of international institutions. International Organization, 55(4), 761-799.

7 See Streeck’s latest book, Streeck, W. (2024). Taking Back Control?: States and State Systems After Globalism. Verso Books.

8 AI Now Institute. (2023). Computational Power and AI. Accessible at https://ainowinstitute.org/publications/compute-and-ai.

9 Kahneman, D., & Tversky, A. (1979). Prospect Theory: An Analysis of Decision under Risk. Econometrica, 47(2), 263-291.

10 See the letter to Science by Lazar, S., & Nelson, A. (2023). AI safety on whose terms?. Science, 381(6654), 138-138.

11 Cf. Raymond, M., & DeNardis, L. (2015). Multistakeholderism: Anatomy of an inchoate global institution. International Theory, 7(3), 572-616.

12 https://www.bloodinthemachine.com/p/ai-is-in-its-empire-era

13 O hEigeartaigh, Sean, The Most Dangerous Fiction: The Rhetoric and Reality of the AI Race (May 25, 2025). Available at SSRN: https://ssrn.com/abstract=5278644.

14 Carlier, A., & Dafoe, A. (2020). Emerging Institutions for AI Governance: AI Governance in 2020. Centre for the Governance of AI. Accessible at https://www. governance.ai/research-paper/emerging-institutions-for-ai-governance-ai-governance-in-2020.

15 Forum on Information and Democracy. (2024). A Voluntary Certification Mechanism for Public Interest AI: Exploring the Design and Specifications of Trustworthy Global Institutions to Govern AI. September 2024. Accessible at https://informationdemocracy.org/wp-content/uploads/2024/09/FID-Public-Interest-AI-Sept-2024. pdf.

16 COM(2025) 165 final.

17 Torreblanca, J. I., & Verdi, L. (2024). Control-Alt-Deliver: A Digital Grand Strategy for the European Union. European Council on Foreign Relations (ECFR). Accessible at https://ecfr.eu/publication/control-alt-deliver-a-digital-grand-strategy-for-the-european-union.

18 Shapiro, J. (2020). Europe’s digital sovereignty: From rulemaker to superpower in the age of US-China rivalry. European Council on Foreign Relations (ECFR). Accesible at https://ecfr.eu/publication/europe digital sovereignty rulemaker superpower age us china rivalry.

19   Cantero Gamito, M (2024). El trilema de la gobernanza: los retos para la democracia global en la era de la IA. Telos, vol. 125, p. 104-108.

STG | Policy Papers Issue | 2025/13

Florence School of Transnational Governance

European University Institute

Via Camillo Cavour, 65a, 50129 Firenze (FI), Italy

Tel. +39 055 4685 545

Email: stg.publications@eui.eu