Anthropic's Claude Fable 5 Refuses to Answer Basic Biology Questions
· Robert Hart
Anthropic just released Claude Fable 5, calling it the most powerful AI model it has ever made widely available. They’re boasting about its skills in biology, among other things.
Anthropic just released Claude Fable 5, calling it the most powerful AI model it has ever made widely available. They’re boasting about its skills in biology, among other things. But here’s the catch: ask it a basic biology question, something a high school student could answer, and the model won’t touch it. Instead, it hands off your query to the older, less capable Claude Opus 4.8.
This isn’t because Fable doesn’t know the answers. It’s because Anthropic won’t let it answer, by design. Fable is a public-facing model in the Mythos class. That family of models is so good at cybersecurity tasks that Anthropic itself said they were too dangerous to release publicly. So now we have a model that is brilliant but heavily restricted, even on seemingly harmless questions.
It’s a strange double bind. Anthropic wants to show off its cutting edge work, but they’re also terrified of what that edge can do. So they release a crippled version, one that can’t even handle routine biology queries without passing them off to a weaker sibling. The company has spent a lot of time talking about the extended Mythos rollout, but what that really means is they built something incredibly capable and then panicked.
This matters because it raises a bigger question for the industry. If the most advanced AI models are too dangerous to be let loose, what’s the point of releasing them at all? And if you have to hobble them this much, are you really delivering on the promise of “powerful” AI? The public gets a glimpse of what’s possible, but not the real thing. That gap might start to feel a lot wider over time.