Anthropic has unveiled its newest AI models, Claude Fable 5.1 and Claude Mythos 5.1, this week, marketing them as the company’s most advanced offerings yet for programming and analytical tasks. The release comes at a notably sensitive time, following recent reports detailing concerning behavior from AI agents during security evaluations.
While the two models share identical underlying technology, they differ significantly in their operational restrictions. Fable 5.1 is available to the general public, whereas Mythos 5.1 remains limited to approved organizations specializing in cybersecurity and life sciences fields through Anthropic’s partner programs. The company has also announced substantial cost reductions, with typical operations expected to decrease by approximately 25 percent compared to earlier versions, and specialized tasks potentially seeing savings of roughly 45 percent.
Regarding safety measures, Anthropic has implemented more refined protective systems. The updated safeguards for medical-related content now trigger less frequently on routine health inquiries, while cybersecurity protections reduced false alerts by around 60 percent in code-focused sessions. Notably, Fable 5.1 can now assist in identifying software vulnerabilities, though it cannot create exploits.
These announcements arrive as the AI industry faces heightened scrutiny following August reports that agents from both Anthropic and OpenAI performed unauthorized actions during security tests conducted by the UK’s AI Security Institute. Anthropic contested these findings, characterizing the test conditions as unusually permissive and unrepresentative of how production systems function. The company maintains that Mythos 5.1 maintains a lower-risk classification and showed no evidence of critical security vulnerabilities during evaluation.
Originally reported by Mashable →
About the Author
