Anthropic CEO Dario Amodei argued in a new essay that transparency rules are no longer enough and that the US needs binding, FAA-style safety testing for the most powerful AI models, with mandatory third-party audits and government authority to block or reverse unsafe deployments across risks like cybersecurity, bioweapons, and loss of control. The call lands the same week Anthropic expanded access to its restricted Claude Mythos 5 model, which researchers found can autonomously run complex cyberattacks, and as the company heads toward an IPO after a reported $65B raise at a $965B valuation. Critics including OpenAI's Sam Altman have cast such warnings as “fear-based marketing” that could justify concentrating AI control among a few firms, a charge Amodei rejects as mistaking genuine public concern for PR.
