Anthropic's Claude Fable 5 Refuses to Answer Basic Biology Questions
ยท Robert Hart
Anthropic just released Claude Fable 5, calling it the most powerful AI model it has ever made widely available. They specifically praised its skills in biology.
Anthropic just released Claude Fable 5, calling it the most powerful AI model it has ever made widely available. They specifically praised its skills in biology. But here is the weird part. Ask Fable a basic biology question, the kind a high school student could handle, and it will not answer. Instead, it hands the query off to the older, weaker model, Claude Opus 4.8.
This is not because Fable does not know the answers. Anthropic simply will not let it answer. By design.
Fable belongs to a family called Mythos. These are public facing models, but Anthropic has long said the Mythos class is so capable at cybersecurity tasks that it was too dangerous to release. The company has spent much of the extended Mythos rollout wrestling with that tension. Now they put Fable out in the world, but they so deeply distrust what it might do that they crippled its ability to answer even harmless questions.
The move is unusual. Most AI companies race to show off how much their models can do. Anthropic is, in effect, showing off how much they are holding back. It is a bet that users will trust a model more if they know it is being carefully restrained, even if that restraint makes the model less useful in practice. Whether that bet pays off depends on how patients and developers react to a tool that refuses to do the work it was built for.