Overview
- Anthropic released Claude Fable 5 to the public on Tuesday, June 9, as its most capable consumer model but with built-in limits on sensitive queries.
- When Fable 5 receives requests about cybersecurity or biological risks, Anthropic redirects those queries to the lower-capability Claude Opus 4.8 using real-time filters.
- Anthropic made an unconstrained version, Claude Mythos 5, available to roughly 200 vetted organizations through Project Glasswing after expanding the program by about 150 partners.
- The company says it ran more than 1,000 hours of internal and bounty adversarial testing and did not find a ‘universal bypass’ that defeats its safety controls.
- Anthropic has filed confidential IPO paperwork with the SEC and faces a White House evaluation, raising questions about how safety, market competition, and infrastructure risk will shape broader access to powerful AI.