Anthropic has released two distinct versions of its latest AI model: Claude Fable 5, available to the general public, and Claude Mythos 5, reserved for what the company calls trusted organizations, according to Wired.
The split is intentional. Fable 5 comes with significant restrictions built in—the model refuses to answer queries related to cybersecurity, biology, and chemistry, topics Anthropic has deemed too dangerous for open access, according to Ars Technica. Mythos 5, by contrast, is the fuller-capability version, offered to vetted "cyber partners."
The rollout hasn't been smooth. After users discovered that Fable 5's guardrails were operating silently—blocking queries without clearly explaining why—Anthropic was forced to make those restrictions visible, according to Android Headlines. The backlash underscores a persistent tension in AI deployment: companies want to limit harm, but users expect transparency about what a model will and won't do.
Anthropics has also released benchmark data comparing Fable 5 against rivals including OpenAI's GPT 5.5 and Google's Gemini 3.5 Pro, as well as its own Mythos Preview, across categories such as coding, reasoning, vision, and cybersecurity, according to Moneycontrol.
The two-tier strategy reflects a broader industry reckoning over dual-use AI—tools powerful enough to serve defensive security researchers and bad actors alike. How well those guardrails hold, and whether users will accept a model that doesn't fully disclose its own limits, is the real test ahead.