The chief executive of Anthropic, Dario Amodei, has issued a stark warning that governments must be empowered to block the deployment of artificial intelligence models deemed to pose unacceptable risks, marking one of the most forceful calls for regulation from a leading AI developer. In a policy essay published this week, Amodei argued that AI systems should undergo mandatory third-party testing to assess threats across categories including cybersecurity and the potential to facilitate biological weapons. If such evaluations conclude that a model presents unacceptable dangers, he wrote, “the government should have the authority to block or discourage its deployment.” The proposal, viewed from Washington as a significant shift in industry posture, comes amid growing global debate over how to govern a technology advancing faster than regulatory frameworks can adapt.

Yet Amodei’s intervention extends beyond safety to a more uncomfortable question: the future of work. In the same essay, he warned that there is a “decent possibility” that AI could cause “significant enduring job loss,” and that this may be “an intrinsic property of the technology and the way it broadly replicates human cognition.” This reframes the debate over automation, suggesting that job displacement is not merely a temporary disruption or a result of corporate short-termism, but a structural feature of AI’s ability to perform cognitive tasks. Analysts in London note that this admission, coming from a CEO whose company builds the very systems in question, lends weight to fears that entire professional sectors could be permanently reshaped.

The dual emphasis on safety and labour disruption reflects a broader tension within the AI industry. While some executives have downplayed job losses as a transitional challenge, Amodei’s essay acknowledges the possibility of a more fundamental shift. He proposes redistributing the economic gains from AI to cushion the blow, though he offers few specifics. From a European perspective, where regulators are already drafting the world’s first comprehensive AI law, Amodei’s call for government veto power aligns with calls for precautionary oversight, but his labour market analysis raises questions about whether existing social safety nets are adequate for the scale of change envisioned.

As the debate moves from boardrooms to parliaments, Amodei’s essay may serve as a watershed moment. By advocating for both stringent safety checks and a realistic appraisal of job displacement, he has placed Anthropic at the centre of a conversation that the industry has often avoided. The coming months will test whether governments are willing to assume the powers he proposes, and whether the public is prepared for the trade-offs that a transformative AI era demands.