Overview
- Anthropic made Claude Fable 5 generally available on Tuesday as a public version of its high-capability Mythos architecture intended for broad use with extra safety layers.
- Fable 5 uses independent classifiers to flag high-risk queries and automatically routes those requests to the less-capable Claude Opus 4.8 as a protective fallback.
- Anthropic requires prompts and outputs to be retained for up to 30 days with flagged material held longer which triggered legal and compliance reviews and has led Microsoft to block employee access while still offering the model to some customers.
- A researcher calling themselves Pliny the Liberator posted claimed multi-agent jailbreaks and an alleged leaked system prompt but Anthropic disputed those posts as not showing a true bypass and apologized for overly aggressive fallbacks while saying it will make refusals and fallbacks visible to users and API clients.
- Security experts say frontier models like Mythos and Fable compress the time to find and weaponize vulnerabilities, which pushes firms to adopt continuous, AI-aware security practices and a tiered access model that keeps the most sensitive versions for vetted partners.