Inside the fight over Claude Mythos 5

Key Points:

• Anthropic was ordered by the Trump administration to suspend access to its advanced AI models, Mythos 5 and Fable 5, for all foreign nationals, including foreign employees, due to concerns about potential security vulnerabilities and "jailbreaking" of safeguards.
• The directive came with a 90-minute ultimatum on Friday, prompting Anthropic executives, including CEO Dario Amodei, to engage in urgent negotiations with multiple US government officials over the weekend to try to reverse the decision.
• The US government’s concerns were reportedly triggered by a credible partner revealing a method to bypass Fable 5’s guardrails, with some reports suggesting involvement of Amazon researchers and fears over China-linked access, though Anthropic disputed the uniqueness of the vulnerability.
• The move has sparked criticism from cybersecurity experts and industry leaders who warn that restricting Anthropic’s models could hinder US AI competitiveness against China and drive international demand toward non-American AI systems.
• Talks between Anthropic and the administration remain ongoing without resolution, highlighting challenges in regulating powerful AI technologies and raising broader questions about access, export controls, and the future of AI governance in the US.