Anthropic proposes safety plan to pace frontier AI development

Anthropic co-founder Dario Amodei has proposed a three-step plan to pace the development of frontier artificial intelligence through independent safety audits and international coordination. Amodei warns that at the current rate of acceleration, autonomous systems could potentially take over the internet within six months to a year, potentially causing hundreds of billions of dollars in damage.

The proposed plan includes granting third-party evaluators, such as METR, permanent employee-level access to systems to verify safety standards. Anthropic has committed to this step unilaterally. The plan also calls for democratic nations to coordinate industry rules and suggests that authoritarian states, including China, participate in shared protocols, such as bans on using AI to develop biological weapons.

While the proposal aims to mitigate existential risks, some critics question the feasibility of a truly neutral watchdog. Concerns have been raised that such oversight could concentrate power among a small group of firms and political actors rather than protecting the public, potentially dulling the discipline provided by market competition and existing liability laws.