Anthropic’s Mythos mess is only getting worse

Key Points:

• Anthropic’s Mythos-class AI models have been offline for two weeks following a Trump administration export control order restricting foreign national access due to security concerns, with no resolution yet in sight.
• The export control process is complicated by a lack of clear frameworks for AI, and negotiations have been slow despite Anthropic’s assertion that its models are safe; a recently flagged vulnerability in Fable 5’s guardrails has intensified scrutiny.
• Security expert Katie Moussouris argues that the reported vulnerability is overblown and reflects essential AI capabilities for defensive security rather than a guardrail bypass warranting severe government action.
• The shutdown has severely impacted Anthropic’s revenue prospects and public image ahead of its IPO, while also creating uncertainty in the US AI industry amid rising global competition, particularly from China.
• The Trump administration’s move marks a significant regulatory intervention in AI, contrasting with its prior deregulatory stance, and has raised concerns among cybersecurity leaders about the approach to AI oversight.