Anthropic has released Claude Fable 5, the general-use version of the same underlying model as Claude Mythos 5.

The model includes safeguards that route some requests to an older system. Anthropic said queries on cybersecurity, biology, chemistry, or distillation are instead handled by Claude Opus 4.8. The company said this fallback triggers in less than 5% of sessions.

Anthropic said the distillation filter was a response to earlier model-extraction attempts. It also identified industrial-scale campaigns by Chinese AI firms DeepSeek, Moonshot AI, and MiniMax.

The company said the activity involved about 24,000 fraudulent accounts and more than 16 million exchanges with Claude. Anthropic said it does not offer commercial access in China.

Researchers criticized Anthropic for initially failing to disclose when some requests were routed to an older system.

The company had initially downgraded some flagged requests without notifying users.

After backlash, Anthropic said on X that users should be able to see when safeguards apply and apologized for “not getting the balance right.”

🔗 Source: South China Morning Post

Recent Anthropic developments