Anthropic opens Claude Fable 5 after locking down Mythos risks
What's new: The model adds guardrails on cybersecurity and biology and scored about 5% above Claude Opus 4.8 in Vals AI tests.
Anthropic has released Claude Fable 5, a public-facing AI model built from its more powerful Mythos system after holding back Mythos over fears it could help hackers breach computer networks. The San Francisco company said Fable 5 uses older technology for part of its work and adds guardrails to block responses tied to cybersecurity, biology and other sensitive areas. Anthropic previously shared Mythos with about 40 organizations that run critical infrastructure to help patch vulnerabilities, and said in June it would extend access to 150 more groups. Vals AI said Fable 5 outperformed other publicly available AI systems overall and ranked roughly 5% above Claude Opus 4.8, though it costs twice as much.