Anthropic Is Still at Odds With the White House Over Claude Fable 5

Key Points:

• The Trump administration concluded talks with Anthropic without lifting export controls on its advanced AI model Claude Fable 5, citing concerns that its guardrails could be disabled to access more powerful capabilities of the Mythos model.
• Anthropic disputes the administration's concerns, arguing that the risks are overstated, while some cybersecurity researchers support this view and claim the export controls may harm AI security efforts and market stability.
• The administration’s restrictions followed alerts from Amazon CEO Andy Jassy and an NSA assessment confirming vulnerabilities that could strip Fable 5’s safeguards, prompting emergency discussions involving Commerce Department officials and Anthropic executives.
• The core disagreement centers on the severity of the jailbreaking risk, with experts noting that guardrails act as speed bumps rather than absolute security boundaries, and that restricting models like Fable 5 may be a temporary measure.
• The incident highlights broader implications for AI labs, signaling that the US government expects early notification and cooperation on advanced model releases, potentially influencing future AI development and deployment strategies.