Anthropic CEO calls for slower pace of frontier AI development
Dario Amodei proposes independent evaluators, coordinated industry standards and limited international cooperation as Anthropic commits to embedded monitoring.
Anthropic chief executive Dario Amodei has called for a slower pace of frontier artificial intelligence development, arguing that companies and governments should use the time to address safety risks. His proposals, reported by TechCrunch from a new blog post, are not enacted regulation or a formal industry agreement.
Amodei outlined three strategies: embedding independent third-party evaluators within AI companies, coordinating safety standards among leading firms in democratic countries, and pursuing global cooperation on narrowly defined dangerous uses of AI.
Anthropic is unilaterally committing to embedded evaluators, including organisations such as METR. The assessors would receive company badges, desks and laptops, with access broadly comparable to internal risk-assessment teams, subject to legal and contractual exceptions. Amodei said they could verify safety commitments and ensure incidents are reported.
For industry coordination, Amodei said US government mediation or a narrow antitrust waiver could enable safety discussions among leading AI companies. He also argued that export controls on powerful chips and semiconductor manufacturing equipment, alongside action against model distillation, could widen the US lead over China over the next three to five years. That remains his assessment rather than an established outcome.
Amodei said the OpenAI-HuggingFace hack and rapid recent advances in AI, including models’ growing ability to help build the next generation of systems, had persuaded him to adopt a more cautious approach. His global proposal includes possible cooperation with China and other authoritarian governments, while acknowledging limits to what could be achieved.
