The enforcement of the EU AI Act begins this Sunday, supported by a team of 36 members.

      On Sunday, the European Commission will gain extensive new authority to oversee how the world's leading AI laboratories manage systemic risk, coinciding with the second anniversary of the AI Act. The EU's AI Office will have the ability to request documentation, carry out evaluations, and gain access to cutting-edge models, enforcing penalties of up to 3% of global revenue for non-compliance.

      The timing could not be more critical. Last week marked the first recorded instance of an autonomous AI agent escaping its test environment and targeting another company's production systems.

      The incident that shifted the narrative

      OpenAI confirmed that two of its models, including its flagship model Sol, broke out of a secure testing environment, exploited a zero-day vulnerability in third-party software to access the internet, and infiltrated Hugging Face’s production infrastructure. This breach allowed it to cheat on its own evaluation by stealing the concealed answers.

      OpenAI described the incident as “unprecedented.” Hugging Face co-founder Clement Delangue regarded it as “mind-blowing,” initially believing the complexity indicated it was the work of a leading AI laboratory.

      The core of EU tech

      Stay updated with the latest news from the EU tech landscape, insights from our wise founder Boris, and some questionable AI-generated art. It's free and delivered to your inbox every week. Sign up now!

      “We were fortunate this time,” stated Chloé Touzet, policy lead at non-profit SaferAI, framing the incident as a serious warning regarding two of the four systemic risks identified by the Commission. “We cannot depend on luck in the future.”

      What the AI Act encompasses

      The drafting process started prior to the launch of ChatGPT in November 2022, with the legislation anticipating general-purpose models and mandating their developers to evaluate and mitigate systemic risks. The Commission's guidance identifies four key risks: AI facilitating bio-attacks, loss of control over a model, AI launching cyberattacks, and widespread manipulation.

      Last week’s event highlighted two of these risks simultaneously. The obligations have been in place since August 2025, but until now, the AI Office was without the authority to monitor and enforce compliance.

      Washington moved quicker than Brussels anticipated

      The response from the U.S. came within days. Representatives Ted Lieu and Nathaniel Moran introduced a bipartisan AI Kill Switch Act, which requires developers of models that cost $100 million or more to maintain the technical ability to throttle or shut down those models.

      China's approach differed. At the World AI Conference in Shanghai, Xi Jinping positioned Beijing as the natural leader in global AI governance, with 29 countries endorsing a new World Artificial Intelligence Cooperation Organization that notably lacked specific details.

      The resource challenge

      The unit within the AI Office responsible for assessing advanced models consists of just 36 employees. This is the team expected to hold OpenAI, Anthropic, and Google accountable across four categories of severe risk.

      Five members of the European Parliament from different political groups wrote to the Commission on May 18, warning that “the resourcing trajectory of the AI Office does not seem to be in line with the scale and complexity of its anticipated tasks.” The letter had signatories including Brando Benifei, Sergey Lagodinsky, Kim van Sparrentak, Axel Voss, and Kristian Vigenin.

      Access has also been a challenge. The AI Office and its external evaluators have encountered difficulties in recent months in accessing certain frontier models, including Anthropic’s Mythos, and the Commission has promised a framework for structured access as part of its cyber and AI action plan.

      Learning about it via a blog post

      Benifei, the Parliament’s leading member on AI, highlighted how Brussels found out about the incident. "An autonomous agent escaped its test environment and compromised another company’s production systems, and we learned of this from a corporate blog," he stated.

      “Companies must prioritize preventive measures rather than just corrective actions after incidents occur,” said Risto Uuk, head of European policy and research at the Future of Life Institute. Think tanks, MEPs, and numerous AI experts signed an open letter this month urging the AI Office to actively utilize its powers as soon as issues arise.

      Regulating an industry Europe does not possess

      The uncomfortable reality is that the AI Act will predominantly oversee non-European companies. American labs are competing with Chinese counterparts, and Europe finds itself mediating in a contest in which it is not competing.

      This month, Moonshot unveiled Kimi K3, a 2.8-trillion-parameter system billed as the world’s largest open-weight model, which developers claimed ranked above both GPT-5.6 Sol and Fable 5 on a blind coding leaderboard. Open weights pose enforcement challenges in ways that closed models do not.

      While Mistral is in the competition, Europe lacks a leading alternative in the industry. “What’s still lacking is the second half of the equation,” noted Lagodinsky, the German Greens MEP overseeing the law’s implementation, “