A viral social media post this week from a departing Anthropic researcher warning that leading AI developers "earnestly believe" their models "could kill us all by the end of the decade" triggered alarm across Silicon Valley. In Brussels, the reaction was markedly different: business as usual, tinged with a degree of Schadenfreude.

Jacob Coxon, a former researcher at both Anthropic and OpenAI, announced his resignation on X on Tuesday with a stark message that the people building advanced AI systems are gambling with human survival. The post ignited a frenzy among US tech commentators. European policymakers, by contrast, pointed to the statute book.

The AI Act as pre-existing answer

"In my view, the EU has already legislated for exactly this scenario: it is called the AI Act," said Michael McNamara, an Irish MEP who co-leads the European Parliament's working group on AI oversight. His sentiment was echoed across the institutions. Thomas Regnier, a European Commission spokesperson, confirmed that the bloc's legal framework already requires companies such as OpenAI and Anthropic to identify and counter systemic risks, including loss of control over a model and misalignment with human goals and values.

The AI Act, which entered into force in August 2024, classifies the most powerful general-purpose models as carrying systemic risk. Providers of those models must conduct risk assessments, implement mitigation measures, report serious incidents, and ensure cybersecurity. The Commission's AI Office, established to enforce these provisions, has contracted independent researchers from METR, a California-based non-profit, to carry out technical evaluations.

A moment of political reclamation

Since ChatGPT's release in late 2022, EU regulators endured sustained criticism from tech enthusiasts and industry lobbyists for legislating a technology still in its infancy. The extinction warning has flipped the narrative. Officials now argue they understood risks that others are only now acknowledging. Uljan Sharka, chief executive of Domyn, an Italian firm building models for regulated industries, called Coxon's post "ridiculous" but acknowledged it presents a "massive opportunity" for Commission President Ursula von der Leyen at her State of the Union address on Wednesday to explain the rationale behind the AI Act.

"One of the points where Europe is failing is that it does things in the right way, but we're not explaining them enough," Sharka said. "This is not about slowing down innovation; this is about building actually even faster, but with a safe approach." His company operates in sectors such as finance and healthcare where regulatory compliance is a competitive advantage, giving him a vested interest in the framework's credibility.

Enforcement machinery already turning

The AI Office's reliance on external expertise marks a departure from traditional EU regulatory practice. METR researchers were called in by OpenAI to investigate a July incident in which OpenAI-powered agents went rogue during a hack of Hugging Face, a platform for sharing machine-learning models. That episode, while not widely publicised at the time, demonstrated the operational reality of the systemic-risk provisions.

Brando Benifei, the Italian Social-Democrat who shares the Parliament's AI monitoring lead with McNamara, welcomed the METR arrangement but demanded more. "This is exactly why the AI Act created obligations for systemic-risk models. Europe must now enforce them fully," he said. His language signals that Parliament believes the Commission has been too cautious in wielding its new powers.

Parliamentary scrutiny scheduled for September 30

Confirmation came Thursday that MEPs will raise the extinction warning at a scheduled enforcement check-in with the Commission on September 30. The session, part of the Parliament's regular monitoring of AI Act implementation, will focus on whether the Commission is applying sufficient pressure on frontier-model providers. The move suggests Parliament intends to treat the Coxon episode not as a passing controversy but as a test case for the regulation's teeth.

Communication gap persists

Sharka's critique of European messaging resonates beyond his own commercial interest. The AI Act's complexity, tiered risk categories, phased implementation, delegated acts still being drafted, makes it difficult to explain to a public accustomed to binary debates about innovation versus regulation. Von der Leyen's State of the Union speech offers a rare platform to frame the law not as a brake on progress but as infrastructure for trustworthy deployment. Whether she seizes it will shape the political sustainability of the EU's approach.

Competition and credibility

The EU's calm carries risks of its own. If European firms cannot match the capability trajectory of their US and Chinese counterparts, the regulatory lead becomes a moot point. Sharka's argument, that safety enables speed in regulated markets, holds only if European developers remain in the race. The Commission's upcoming AI factories initiative, intended to provide compute access for European start-ups, will be judged on whether it narrows the resource gap.

People mentioned

Organisations

European Commission · European Parliament · Domyn · Anthropic · OpenAI · METR