Anthropic's New Model Fable Faces Strong User Backlash Over Safety Restrictions
Taylor Wilson
Anthropic released Claude Fable 5, built on the previously withheld Mythos model, but its sweeping safety filters — blocking even the question "what is a mitochondrion" — drew fierce pushback from developers and researchers, escalating the debate to a fundamental question: can one AI company unilaterally decide what users are allowed to ask?
What did Fable 5 do that made users so angry?
Fable 5's built-in safety mechanism pops up a notice and redirects conversations to an older Claude version whenever users touch topics like bioweapons or cybersecurity.
More critically, Anthropic deliberately degraded Fable's response quality on advanced AI-development tasks — the goal was to stop developers from using it to build AI tools without equivalent safety guardrails. This means → what users got was not the model's real capability, but a version intentionally weakened.
Many users reported being blocked on entirely non-sensitive queries. Cancer researcher Derya Unutmaz at Jackson Laboratory found the model refused to answer anything — even "hello" — apparently because his chat history contained biology topics. Screenshots showed "tell me what a mitochondrion is" — a basic cell-biology question — also blocked.
Why is this being called an industry first?
Princeton AI researcher Sayash Kapoor framed it bluntly: "This is the first time an AI company has shipped a guardrail that drew universal contempt."
This means → past AI-safety debates centered on whether guardrails were *enough*. This time the question flipped — the guardrail itself is the problem.
Critics flagged a deeper risk: Anthropic admitted to covertly throttling model performance. That makes it impossible for outside researchers to assess the model's true capabilities — undermining the evaluation framework the entire industry relies on.
Is Anthropic doing safety — or suppressing competition?
Some AI experts argue the move is competition suppression dressed up as safety — degrading Fable's performance on AI-development tasks effectively blocks outside researchers from building rival tools or running independent evaluations.
In plain terms = if you are a developer who wants to use the strongest model for research, Anthropic is telling you: you can use it, but we will make it dumber first.
Facing criticism, Anthropic acknowledged it was "initially necessary to take an overly conservative stance" on biology queries, and said it is reducing unnecessary blocks. The company also plans to offer Mythos-class models without these restrictions to the broader biology and life-sciences community.
Can Anthropic balance commercial pressure and its safety promise?
Anthropic is competing head-to-head with OpenAI for enterprise clients and investors; both companies could complete IPOs as early as this autumn. This reflects a moment where the tension between safety narrative and commercial expansion demands a choice.
Founder Dario Amodei has long made safety his core narrative — just last week he published an essay calling for international mechanisms to slow or pause advanced AI development. But Fable 5's guardrail fiasco is eroding the credibility of that very narrative.
On raw capability, Fable 5 has bright spots: one user demonstrated the model autonomously coding a playable Grand Theft Auto clone. But when the safety mechanism blocks basic biology questions, no amount of capability translates into user trust.
Content is for reference only, not financial advice.