Anthropic Launches Fable 5 Model, Blocks Cybersecurity-Related Capabilities

Odaily Odaily reports that Anthropic has released the Mythos version of Fable 5, which lacks the ability to execute cybersecurity tasks. The model is equipped with safety guardrails to prevent it from responding to specific queries, including those related to cybersecurity and biology. In these cases, its Claude chatbot will route responses through a different model named Opus 4.8.

Anthropic has also released a new version, Mythos 5, which does not include certain safety measures. This version will be provided to selected groups through a project named Project Glasswing.

Fable 5 is designed to improve capabilities in coding and other specialized tasks. In testing, payment processing company Stripe completed lengthy software engineering tasks in one day that would have otherwise required a team two months of manual execution. Additionally, Anthropic stated that it ran an external bug bounty program and found no universal jailbreak vulnerabilities in over 1,000 hours of testing. (Bloomberg)