Overview
- Anthropic CEO Dario Amodei published a long policy essay and a legislative proposal on June 10 calling for mandatory third‑party testing of 'frontier' AI and for governments to be able to block or reverse any deployment judged unsafe after independent review.
- Anthropic says findings from its internal Mythos model, which identified thousands of software flaws and other cybersecurity vulnerabilities during closed testing, show that advanced models already create concrete national‑security and infrastructure risks.
- The company paired its safety demands with an economic plan that includes wage insurance, retention incentives, possible capital‑account ideas and a $200 million commitment to fund research on AI’s societal impacts.
- Anthropic is in active conflict with the U.S. government after the Pentagon labeled the company a supply‑chain risk and banned Claude, prompting Anthropic to sue the administration while the White House maintains a separate voluntary 30‑day review process and Congress has not yet moved toward binding law.
- Industry reaction is split: senior Microsoft executives warned against anthropomorphizing models and criticized parts of Anthropic’s rhetoric, while analysts warn that mandatory audits and enforcement would raise verification, cost and geopolitical challenges and could advantage large incumbents.