Anthropic CEO Dario Amodei published an essay on September 12 proposing a framework to align AI safety with development speeds. The core of his plan involves requesting an antitrust waiver from the US government, which would allow competing AI firms to collaborate on safety protocols without legal risks.
The proposal includes three main components: integrating independent third-party evaluators into AI labs, creating coordinated safety standards among democratic nations, and establishing dialogue channels with authoritarian states like China.
FTC Chair Andrew Ferguson expressed skepticism on September 15, noting that antitrust exemptions for major AI companies could create barriers for new market entrants and solidify the position of current incumbents.
Industry leaders have responded with mixed reactions. OpenAI CEO Sam Altman supported the use of third-party evaluators on September 13, but did not endorse the antitrust waiver. Nvidia CEO Jensen Huang stated on September 15 that current laws are sufficient for safety efforts.
Amodei’s proposal is driven by concerns regarding the rapid development of AI models, specifically the potential for recursive self-improvement. He warns that significant risks from these advancements could emerge within the next six to twelve months.
