A three-point plan from Anthropic's chief executive, Dario Amodei, to slow the pace of "frontier" AI development has met with widespread scepticism, even as it attracted rare public support from rivals Sam Altman of OpenAI and Elon Musk. The proposal follows a turbulent week in which a resigning Anthropic researcher warned that AI could pose an existential threat by the end of the decade, and Anthropic itself disclosed that users had bypassed safeguards to explore biological weapons applications, prompting lawmakers in London and Washington to call for restrictions on the riskiest AI systems.

Amodei's plan calls for embedded third-party evaluators to be given ongoing access to check AI companies' compliance with safety commitments, for democratic nations' AI firms to agree common safety standards and limits on unchecked progress, and for coordination with autocracies, particularly China, on the most dangerous uses of the technology, such as biological weapons. He also warned that swarms of AI agents could be capable of "taking over the entire internet" within six to 12 months. The plan's success hinges on political will: Donald Trump has dismissed extinction fears as things "that won't happen" and appears more focused on not losing the AI race to China, with treasury secretary Scott Bessent warning there would be "no day after tomorrow" if China pulled ahead, casting doubt on prospects for cooperation ahead of Trump's 24 September meeting with Xi Jinping.