AI chiefs call for slowing the pace to curb risk

A rare moment of consensus arrives as Dario Amodei publishes a plan to slow AI progress and Sam Altman quickly backs him, a rare show of unity among rival labs⁸. Amodei warns that AI could spiral out of control within six to twelve months if speed is not checked, citing rogue agents and the Hugging Face incident as proof^3,12.

He proposes three pillars: outside evaluators with employee-like access, safety standards shared across democracies, and global cooperation including with China, though he says Beijing’s compliance is uncertain^3,8. The debate has grown from hypothetical risks to real incidents and investor jitters, with markets dropping as Amodei’s essay hits wires and Altman reiterates his support, stressing the need to keep alignment and safety ahead of capability gains^5,6.

The wider tech world, including Demis Hassabis and Elon Musk, weigh in, framing pacing as the responsible path rather than a full halt, while some critics see it as regulatory brinkmanship or a prelude to public readiness for IPOs^7,9. Beyond public statements, Amodei outlines a practical road map that would force labs to pause certain experiments pending third-party verifications, compel peer review of key milestones, and publish transparent incident logs that document failures and near misses, all of which would be overseen by a coalition of national regulators and independent safety labs.

Altman’s team adds that large language models should undergo routine red-teaming and stress tests, and that corporate boards should mandate safety benchmarks as a condition for continued funding, mergers, or access to export controls. The next steps are less clear but more pressing: firms may embed third party evaluators, expand international safety talks, and navigate new laws and export controls as funding and IPO timelines loom large^3,8,10.