Claude Fable refuses to answer biology questions — and that's deliberate

Anthropic has released Claude Fable 5, positioning it as the most powerful model ever made available to the general public. But this seemingly ideal model has an unusual limitation: it refuses to answer basic biology questions and redirects them to an older, more conservative model, Claude Opus 4.8.

The paradox of a powerful model with limits

The restriction looks odd. How can the most powerful model fail to answer a question an ordinary schoolchild could handle? Anthropic says it is not because Fable doesn't know the answer. The company deliberately forbade it. The model has built-in mechanisms that recognize certain types of biology questions and automatically redirect them to Claude Opus 4.8.

This is neither an accident nor a bug — it is an intentional part of the model's architecture. When a user asks Fable about the structure of a cell or the mechanism of photosynthesis, they get a suggestion to turn to Opus instead of a direct answer.

Mythos: too powerful for full access

Fable belongs to the Mythos family — a group of models that Anthropic refused for a long time to release to the public. By the company's assessment, these models are so good at cybersecurity tasks that they pose a potential risk. They find vulnerabilities too easily and could be used for malicious purposes.

Despite these concerns, market pressure and user demand pushed Anthropic to release Fable 5. The company did not want to take the full risk, however. Instead, it added a series of restrictions and safety checks.

• The model is exceptionally strong in cybersecurity
• Anthropic feared fully open access
• Fable 5 ships with built-in restrictions
• Biology questions are only one example of such restrictions

This is the first time Anthropic has openly released a Mythos-class model in a restricted form. The decision reflects the company's attempt to balance innovation and responsibility.

How it works in practice

The system runs on relatively simple logic. When a user asks a biology question, Fable recognizes it and does not generate an answer. Instead, it suggests switching to Claude Opus 4.8. It is not quite a block: the user can still get the information they need, but through a different model.

The approach resembles intermediate checks in critical systems. Anthropic is, in effect, telling users: "We are releasing a powerful model, but in certain areas we will steer you to a more proven option." It combines innovation with conservative risk management.

What this means for the AI industry

Fable's restriction shows a growing dilemma in AI: how do you give people access to powerful technology without creating risks? It is not a simple question, and there is no clear-cut answer. Anthropic chose a path of partial openness — release a powerful model, but with built-in "brakes" on certain capabilities.

This approach could become a template for other companies. When they release their own powerful models (and Anthropic's competitors are already working on them), they will likely add similar restrictions. This could become the new norm: releasing strong models with built-in conservatism.