Anthropic has released two updated models this month, Claude Mythos 5.1 and Claude Fable 5.1, both built on what the company calls its most capable underlying model to date, aimed particularly at coding and complex knowledge work.
Two models, one underlying system
The split between the two names reflects Anthropic's approach to handling especially sensitive capabilities. Fable 5.1 is the version available for general use, with safety filters in place for topics like cybersecurity, biology, chemistry and model distillation — if a request touches on those areas, it gets automatically handled by a less capable model instead. Mythos 5.1, built on the same underlying system, has those particular safeguards lifted, but access is limited to a small number of trusted organizations doing legitimate security research, such as scanning codebases for vulnerabilities.
Where it's already showing up
Anthropic says its own vulnerability-scanning product, Claude Security, is now powered by Mythos 5.1, and the model is also being used inside Claude Code, the company's coding assistant, which received a broader update around the same time.
A wider industry pattern
Anthropic isn't alone in taking this split-access approach. Alongside its own release, the company also joined more than 190 other AI providers this year in signing onto an EU transparency code that requires certain AI-generated text to carry an invisible digital watermark — a small but telling sign of how much of frontier AI development now happens under active regulatory attention, not just competitive pressure from rivals like OpenAI and Google.
Reporting on this story also appeared at reconnAI, which has more technical detail if you want to go deeper.