Anthropic Fable 5: Guardrailed AI for Public Use

Anthropic has released a new version of its Mythos model, called Fable 5, for public use. The company states Fable 5 includes guardrails to prevent it from answering questions on cybersecurity and biology, areas that made the original Mythos too dangerous for public release. What's interesting is Anthropic extensively tested Fable 5 with hackers, and none were able to bypass its safeguards. When attempts were made, Anthropic's less powerful Opus 4.8 model provided answers instead. Anthropic acknowledges that without safeguards, Fable 5 could be misused, especially in finding and exploiting software vulnerabilities, which could lower the cost of cyberattacks. Customers in early testing reported Fable 5 significantly reduced software publishing time and performed well on reasoning tasks. The company also rolled out an upgraded Mythos 5 to select customers, which it claims has the strongest cybersecurity capabilities of any model globally. Both Fable 5 and Mythos 5 are priced lower than the previous Mythos version, but remain more expensive than Anthropic's other models due to analytical tasks. This matters because it shows a powerful AI model is now publicly available, but with critical safety measures in place.