The European Union’s AI Act, which entered into force in August 2024, represents the world’s first comprehensive regulatory framework for AI. Implementation falls to the European AI Office, established to develop technical standards for applying the law, coordinate enforcement across 27 member states, and provide guidance to companies navigating the law’s requirements. For the AI Act to succeed, the AI Office must provide entrepreneurs with regulatory predictability and an assurance that requests for assistance will be met with competence and without delay. To make this an even more formidable challenge, the AI Office needs to do this with a budget of only €46.5 million. Composed of around 125 staff members, the entire institution runs on less than 50 percent of what a single researcher has been offered to work at Meta.¹

The AI Office also needs to carefully navigate the various bureaucratic rules and procedures that characterise Brussels and the institutions that call it home. To give just one example, officials at the Office need to be selected with “equal representation” of member states in mind, and they need to pass a standardized test that does not always reflect the skills and competencies needed to succeed in the role. Such hiring procedures are perhaps admirable from a point of view that emphasizes diversity, objectivity and fairness, but they directly hamper the AI Office’s ability to hire from what is realistically a small and geographically concentrated community of AI experts.  

Without the ability to hire the right people and to invest in technical competence, the AI Office will find it difficult to quickly respond to companies’ concerns, provide effective guidance to AI startups, and standardize enforcement measures. Such regulatory uncertainty and dysfunction runs the risk of further cementing Europe’s technological stagnation relative to the United States and China. Failure to properly resource the AI Office threatens to transform Europe’s new legal framework into yet another source of regulatory uncertainty driving AI innovation elsewhere. 

Europe Cannot Afford More Technological Stagnation

The seriousness of the threat of regulatory dysfunction in the AI space becomes clear when viewed against Europe’s broader technology trajectory and the imminent threat of economic stagnation. The link between technology development and economic growth is so strong that once you exclude the tech sector from comparisons of productivity growth between the EU and the United States over the past twenty years, the gap between the two regions disappears almost entirely.² Europe is behind its global competitors in seven out of eight key technologies.³ Out of the world’s top 50 tech companies, only four are European. Only 6 percent of total global funding for AI start-ups goes to EU-based companies, compared to 61 percent for the United States and 17 percent for China.⁴ Europe invested a mere 11 billion in AI in 2023, compared to 67 billion in the United States.⁵ Most tellingly, between 2008 and 2021, almost a third of the “unicorns” founded in Europe left the continent, most heading to America.⁶ Startups are 40 percent more likely to secure funding within their first five years in the United States compared to Europe.⁷

These investors and entrepreneurs are fleeing regulatory uncertainty, administrative complexity, and the inability to get clear, timely guidance from under-resourced government agencies. 

The effects go beyond economics. Prosperity is power, and as the former CEO of Google put it, the “nexus between technological innovation and global domination dates back centuries.”⁸ Even before the tech boom and the decoupling of growth rates across the Atlantic, Europe was predicted to account for only 9 percent of global GDP by 2050, whereas it currently accounts for 15 percent.⁹ “Europe is finished,” say some.¹⁰ Joschka Fischer – Germany’s former foreign minister and vice chancellor – has asked whether Europe will be “the world’s biggest loser.”¹¹

At the same time, Europe needs technological strength to maintain strategic autonomy. Between 75 percent and 90 percent of global wafer fabrication capacity is concentrated in Asia, leading the European Commission to openly acknowledge that if “the EU does not act, we risk being vulnerable to coercion.”¹² Modern defense requires advanced AI for intelligence analysis, semiconductors for weapons systems, quantum computing for secure communications, and autonomous systems for battlefield awareness.  

A possible conclusion one might draw when looking at this data is that Europe is repeating mistakes of the past: by overregulating technology, they are scaring away innovators and entrepreneurs. At first sight, this conclusion is cogent – complaints about excessive red tape are commonplace in criticisms leveled against the EU.¹³ However, regulatory uncertainty, unpredictability, and inconsistency can be even more poisonous to technological innovation than regulatory excess. Indeed, now that the AI Act is law, it may be precisely the EU’s inability to effectively and fairly implement its regulatory vision that spells doom for Europe’s technological ambition and geopolitical security. Rather than taking an apologetic stance on its state-of-the-art legal framework and thinking about walking it back or “stopping the clock,” the EU should lean into embracing the task of competently and consistently implementing it, thereby creating the conditions most conducive to experimentation and innovation.

Regulatory Uncertainty, Not Regulation Itself, Suffocates Innovation

The one thing that entrepreneurs and investors fear more than properly enforced regulation is improperly enforced regulation. A regulatory environment marked by uncertainty discourages investment and impedes innovation. 

Even small startups can navigate strict requirements if those requirements are clearly communicated and defined. What kills innovation is ambiguity. Not knowing whether your product is compliant, waiting months for regulatory guidance, or watching requirements shift unpredictably presents a significant hurdle to investing in big bets. For decades and across industries, companies have called for more, not less regulation, to avoid legal uncertainty.¹⁴ AI is no exception: from Sam Altman (OpenAI) to Sundar Pichai (Google) and Dario Amodei (Anthropic), industry leaders have been at the forefront of demanding sensible interventions.¹⁵

In contrast, clear policies and well-staffed regulatory bodies to implement them can allow for faster, more consistent guidance to startups and other companies. These bodies can engage proactively with companies to publish clear technical standards and evaluation criteria. They can resolve edge cases and answer novel questions in weeks rather than months. The link between staff capacity and the speed at which a regulator moves is well-established: in America’s federal Food and Drug Administration (FDA), new drug application review times shorten by 3.3 months for every 100 additional staff members.¹⁶ The FDA is not a light-touch regulator by global standards.¹⁷ And yet the United States leads the world in pharmaceutical innovation because the FDA provides predictable, science-based evaluation processes.¹⁸ Companies know what evidence they need to provide, what standards they must meet, and roughly how long approval will take, which allows them to plan development roadmaps, raise capital against milestone timelines, and make strategic decisions with confidence. 

Under-resourced regulators also inadvertently favor incumbents. Large companies can afford large compliance teams, expensive legal counsel, and the runway to absorb regulatory delays. They can navigate ambiguity through lobbying, through sustained engagement with under-staffed agencies, and through their ability to shape standards processes through resource advantages. Startups cannot. It is no coincidence that European legislation is required to obey the principle of “proportionality,” allowing smaller actors to find alternative and easier means through which they can comply with regulation.¹⁹

Well-resourced and technically savvy regulators can level this inherently uneven playing field. They can provide free (or subsidized) guidance to startups, create tailored compliance tools and templates, offer customizable sandboxes where new entrants can test approaches with regulatory feedback (as mandated by Article 57 of the EU AI Act), and ensure that standards are clear enough that any company can navigate them.²⁰

The AI Office Faces Challenges in Recruiting Technical Talent

Launched in 2024 and tasked with the monumental challenge of implementing and enforcing the AI Act in close coordination with member states and their respective authorities, the EU AI Office has had an impressive first year. One of their greatest achievements was crafting and publishing the General-Purpose AI Code of Practice, a set of clear guidelines for how AI companies can cost-effectively and flexibly comply with various provisions in the AI Act. The Code of Practice has been so successful that all frontier AI companies except for Meta have signed it, and OpenAI has gone so far as to send a letter to the Governor of California asking him to recognize adherence to the Code as an equivalent determination of compliance in his state.²¹ The AI Office stands out in the European Union, an infamously bureaucratic monster of an institution, as having created an unusually entrepreneurial and fast-paced institutional culture. This atmosphere has attracted technical talent: the office now employs International Mathematical Olympiad (IMO) medalists, ex-YC founders, and former employees of Google, Microsoft, Meta and RAND.

Even more impressively, the AI Office has done all of this – and many other things that have gotten comparatively less attention – on a shoestring budget of €45 million. While comparisons with other bodies in neighboring jurisdictions are notoriously difficult, it is striking that the UK AI Security Institute (AISI) – which serves a single nation of less than 70 million people rather than 27 member states with a combined population of over 450 million – can leverage approximately twice the resources that are at the disposal of the AI Office.²² To put the AI Office’s budget into perspective, we can compare it against the recent pledges to invest in AI infrastructure as part of the EU Commission’s AI Continent Action Plan. “InvestAI,” which is a crucial component of this plan, leverages a total of €200 billion to boost innovation and ensure Europe’s competitiveness, of which €50 billion will come from public funds.²³

One way the AI Office is feeling this pain is staffing.²⁴ Effective AI governance requires world-class technical expertise, including machine learning researchers who understand frontier capabilities, security experts who can assess dual-use risks, domain specialists who can evaluate sector-specific applications, and legal scholars who can translate technical realities into robust regulatory frameworks. Top candidates are often in high demand and can command premium salaries in the private sector, where seven-figure salaries are not unheard of, an imbalance that is exacerbated by the often-temporary nature of positions at the AI Office.

The procedural hurdles are another challenge: if you want to work at the AI Office permanently, the only way to do so is to subject yourself to an arduous and lengthy hiring process mandated by the European Union, the infamous “Concours.” Compare this situation, again, with the UK’s AISI, which employs over 250 people. Unconstrained by the often strict hiring requirements the European Union has imposed upon itself, and with a greater flexibility to hire foreign workers and pay them higher salaries, the AISI is a case study of how public service can become an enticing and entrepreneurial option that attracts top hires from across the globe.

That said, the AI Office has already grown very fast since its inception, and mindlessly adding staff members to outcompete other regulators on statistics will not be a viable path. On the contrary, scaling too quickly and having to lower the bar could result in a less competent, less nimble, and more and needlessly bureaucratic AI Office.

What Needs to Change?

Improving the resources of the AI Office would be a valuable first step for the EU to credibly signal a commitment to complement legislative innovation with enforcement support. However, throwing more money at a complex and fast-evolving problem will not be a panacea.

Most urgently, the European Union’s standard hiring procedures need to be updated to account for the idiosyncratic demands of an agency tasked with overseeing what might be the most important technology humanity has ever invented, and to account for the rapid pace at which state-of-the-art models improve and evolve.²⁵ To attract the best and brightest, the EU should reconsider its policies when it comes to setting salaries, especially when it comes to roles that are difficult to fill because they require the kind of technical expertise that is also highly valued and compensated for in the private sector. Where developing and procuring expertise in-house proves difficult, regulators can explore alternative methods to facilitate knowledge transfer, like secondments or advisory councils.²⁶ Where expertise cannot be brought in-house because of the nationality of those who could provide it, innovative solutions are required. The UK AISI, for instance, has mobilised up to 15 million GBP for the “Alignment Project” to disburse grants for external researchers, allowing them to stay at their current university or employer while also enabling them to contribute to the mission.²⁷ Closer collaboration with AI companies and exclusive access to the frontier of technology development might also make a job at the AI office more attractive.

There are some early signs that things are moving in the right direction. The AI Office’s mandated scientific panel of independent experts, for instance, allows for a limited number of non-EU nationals to be represented, a measure that significantly expands the pool of possible recruits.

Progress on this reform is slow, though: There remains a strong commitment to geographic equal representation that sometimes creates challenges for the AI Office. To this day, only citizens of EU member states can become officials and work at the European AI office, a move that restricts the available talent pool to just 5.5 percent of the world’s population.²⁸ To make things even more difficult, according to the rules laid out in the announcement of AI Office’s scientific panel, every member state is guaranteed at least one spot on the panel.²⁹ This includes seats for countries like Malta or Cyprus, where AI talent is even more scarce than it already is in countries like Germany, France, or the Netherlands. These larger countries, in turn, can send a maximum of three representatives, even if their concentration of expertise is significantly higher. Drawing from small member states like Malta to recruit experts simultaneously creates a potential novel challenge for these countries, who are required by the AI Act to establish national bodies that liaise with the AI Office to ensure rules are enforced and compliance monitored on the national level.³⁰ Due to this “brain drain” and funding shortfalls, capacity is already at its limits: experts decry some states’ inability to even fund existing data protection agencies, calling them “almost broke.”³¹ As of November 2025, fewer than half of EU member states had dedicated budgets for AI strategies.³²

Hence, efforts to foster the development of regulatory competence need to extend to the national level. This should be complemented by information-sharing agreements between regulatory agencies that help to reduce the workload of any individual body while ensuring that enforcement rules are applied uniformly across the Union. To avoid making the same mistake twice and to emulate best practices instead, there are lessons to be learned from the UK AISI, the FDA, and regulatory entities in other jurisdictions.

Finally, the AI Office would be well-advised to itself leverage the powers of the technology it oversees: beyond the more mundane use cases of assistance with legislative drafts, data analysis, and automated translations, there are potential opportunities to leverage AI for the tasks that are more unique to the AI Office, including horizon scanning for risks or compliance monitoring. The AI Office could provide tools that allow AI developers to self-assess their overall preparedness or the adequacy of their Safety and Security Model Reports (as mandated by the Code of Practice).³³ Products like “Elsa” – an LLM-powered tool to expedite clinical protocol reviews and reduce time for scientific evaluations used by the FDA – have already shown what is possible, and can serve as an inspiration to possible future applications that enhance regulatory efficiency.³⁴

Can the AI Office Achieve Regulatory Excellence?

We must face the facts. It is undeniable that Europe suffers from regulatory complexity and fragmentation. It takes an average of 19 months for new EU laws to pass, and over 50 percent of small and medium sized enterprises flag high regulatory burdens as their greatest challenge.³⁵ To pretend that Europe is a nurturing environment for the daring would be to close one’s eyes to the harsh reality. But the solution to bad regulation is not less regulation – it is better, more efficient regulation. In domains like AI, where risks are legitimate and emergent, expertise is scarce, and public concern is high, well-resourced institutions are needed to create, enforce and update rules that are adaptable, proportionate, and fit for purpose.  

While many have seen the mere passage of the AI Act as a significant achievement, the real battle of enforcement and implementation has only just begun. And for that, it will not be enough to throw hundreds of billions of euros at supercomputers and gigafactories, as is planned through InvestAI and the AI Continent Action Plan.³⁶ These measures need to be complemented by a credible commitment to ease red tape and to invest in the personnel and resources needed to shape the safe and innovation-friendly environment that institutions like the AI Office can nourish. In this manner, Europe can build safe harbors for innovators and technologists to remain in Europe. If the continent fails to do this, the effects will reverberate beyond a few European champions and their bottom lines. It will leave Europe dependent on infrastructure and products developed elsewhere, subject to foreign influence and vulnerable to coercion.³⁷

The question, therefore, is not whether Europe can afford to properly resource its AI governance institutions. The question is whether it can afford not to.  

1. “European AI Office: Shaping Europe’s Digital Future,” European Commission, October 23, 2025; Benj Edwards, “At $250 Million, Top AI Salaries Dwarf Those of the Manhattan Project and the Space Race,” Ars Technica, August 1, 2025. ↩︎
2. “The Future of European Competitiveness, Part A: A Competitiveness Strategy for Europe,” European Commission, September 3, 2025. ↩︎
3. “Study: Europe Is Behind on Almost All Critical Technologies; Current Economic Security Strategy Is Too Defensive,” DIGITALEUROPE, June 20, 2024. ↩︎
4. “The Future of European Competitiveness,” European Commission. ↩︎
5. “AI Index,” Stanford University Human-Centered Artificial Intelligence, accessed December 19, 2025. ↩︎
6. “The Future of European Competitiveness,” European Commission. ↩︎
7. Padraig Nolan, “Across the Pond for Profits: European Startups Head to the US,” Center for European Policy Analysis, June 21, 2024. ↩︎
8. Eric Schmidt, “Innovation Power: Why Technology Will Define the Future of Geopolitics,” Foreign Affairs, February 28, 2023. ↩︎
9. John Hawksworth “The World in 2050: The Long View: How Will the Global Economic Order Change by 2050?,” PwC, February 2017. ↩︎
10. Allister Heath, “Europe Is Finished,” The Telegraph, January 31, 2024. ↩︎
11. Joschka Fischer, “Will Europe Be the World’s Biggest Loser?” Project Syndicate, May 18, 2023. ↩︎
12. “The Future of European Competitiveness,” European Commission. ↩︎
13. Katie Prescott, “Tech Regulation Holding Europe Back, Warns Major Investor,” The Times, November 9, 2025. ↩︎
14. Moritz von Knebel, “The Fight Over Standards in the Chemical Industry – a Chain Reaction?” AI Policy Bulletin, February 27, 2025. ↩︎
15. James Clayton, “Sam Altman: CEO of OpenAI calls for US to regulate artificial intelligence,” BBC, May 17, 2023. ↩︎
16. Daniel Carpenter, Michael Chernew, Dean G. Smith, and A. Mark Fendrick, “Approval Times for New Drugs: Does the Source of Funding for FDA Staff Matter?,” Health Affairs vol. 22 (2003). ↩︎
17. Jennifer Ko, “What the FDA Can Teach Us About Regulatory Excellence,” The Regulatory Review, January 16, 2018. ↩︎
18. Sujai Shivakumar, Jeffrey Depp, and Tisyaketu Sirkar, “Understanding the U.S. Biopharmaceutical Innovation Ecosystem,” Center for Strategic and International Studies, August 15, 2024. ↩︎
19. “Principle of Proportionality,” EUR-Lex, accessed December 19, 2025. ↩︎
20. “Regulation (EU) 2024/1689 of the European Parliament and of the Council on Laying Down Harmonised Rules on Artificial Intelligence (Artificial Intelligence Act),” European Union, Artificial Intelligence Act Article 57, June 13, 2024.  ↩︎
21. Pieter Haeck, “EU Signatories AI Code Practice Amazon Google IBM Microsoft AI,” Politico, August 1, 2025; Christopher Lehane (Chief Global Affairs Officer OpenAI) to Governor Gavin Newsom, “Recognition of international and federal AI safety frameworks for state law compliance,” OpenAI, August 11, 2025. ↩︎
22. Viscount Camrose (Parliamentary Under Secretary of State, Department for Science, Innovation & Technology) to Baroness Stowell, “Large Language Models Committee inquiry,” UK Parliament, December 8, 2023. ↩︎
23. European Commission, “EU launches InvestAI initiative to mobilise €200 billion of investment in artificial intelligence,” February 20, 2025. ↩︎
24. Frank Hersey, “EU AI Office Faces Hiring Challenges, Push for International Collaboration, Tudorache Says,” MLex, April 17, 2024. ↩︎
25. Catherine Clifford, “Google CEO Sundar Pichai: AI Is More Important Than Fire, Electricity,” CNBC, February 1, 2018. ↩︎
26. “Applications Open to Bring 25 Top Tech Minds into Government, to Accelerate AI-Driven Growth and Modernise Public Sector,” UK Government Department for Science, Innovation and Technology, May 7, 2025; “AI Advisory Body,” United Nations, accessed December 19, 2025. ↩︎
27. “The Alignment Project,” UK AI Security Institute, accessed December 19, 2025. ↩︎
28. “Facts and Figures on the European Union,” European Union, accessed December 19, 2025. ↩︎
29. “Commission Seeks Experts for AI Scientific Panel,” European Comission Digital Strategy, accessed December 19, 2025. ↩︎
30. “Article 70: Designation of National Competent Authorities and a Single Point of Contact,” European Commission Regulation (EU) 2024/1689 (Artificial Intelligence Act), accessed December 19, 2025. ↩︎
31. Tonya Riley, “EU AI Act Enforcement Hampered By Lack of Funding, Expertise,” Bloomberg, May 9, 2025. ↩︎
32. Gosia Nikowska and Karine Perset, “Tracking Europe’s Progress on AI: Insights from the Implementation of the EU Coordinated Plan on Artificial Intelligence,” OECD.AI, November 26, 2025. ↩︎
33. “The General-Purpose AI Code of Practice,” European Commission, accessed December 19, 2025. ↩︎
34. “FDA Launches Agency-Wide AI Tool to Optimize Performance for the American People,” U.S. Food and Drug Administration, June 2, 2025. ↩︎
35. “The Future of European Competitiveness,” European Commission. ↩︎
36. “EU launches InvestAI initiative to mobilise €200 billion of investment in artificial intelligence,” European Commission, February 10, 2025; “The AI Continent Action Plan,” European Commission, April 9, 2025. ↩︎
37. “Report on European technological sovereignty and digital infrastructure,” European Parliament, November 6, 2025. ↩︎