Anthropic’s Fable AI refuses basic biology queries despite high claims

Anthropic recently unveiled Claude Fable 5, positioning it as the company’s most advanced publicly available AI model to date. With claims of superior performance in biology and other domains, the model’s debut seemed poised to redefine expectations for large language models. Yet, early testing reveals a puzzling limitation: Fable refuses to answer even the most fundamental biology questions—ones a high school student could easily resolve.

Instead of providing direct responses, Fable redirects users to Claude Opus 4.8, Anthropic’s previous flagship model. This behavior isn’t a reflection of Fable’s knowledge—it’s a deliberate design choice enforced by Anthropic’s safety protocols. The company has labeled Fable as part of its Mythos-class of models, a designation reserved for systems so proficient in cybersecurity tasks that Anthropic deemed them too risky for public release.

Why Fable dodges basic biology questions

At first glance, the restriction appears counterintuitive. If Fable possesses advanced reasoning abilities, why does it avoid straightforward queries? The answer lies in Anthropic’s layered safety framework. The company has implemented safeguards to prevent the model from generating responses that might inadvertently encourage harmful behavior, even when the intent is educational.

For example, a user asking Fable to explain how photosynthesis works might receive a reply like, "I can’t assist with that request." Anthropic has not clarified whether these restrictions apply universally or only to specific topics, leaving users to speculate about the boundaries of the model’s functionality.

The Mythos-class dilemma: power vs. safety

Anthropic’s Mythos series was originally positioned as a breakthrough in AI capabilities, particularly in domains like cybersecurity and complex problem-solving. However, the company’s decision to withhold the most powerful versions from public access underscores a growing tension in the AI industry: balancing innovation with risk mitigation.

While Fable is marketed as a public-facing model, its restrictions suggest a cautious approach to deployment. Critics argue that such limitations could hinder the model’s utility, especially in fields like education where basic scientific queries are routine. Meanwhile, Anthropic maintains that these safeguards are necessary to prevent misuse, even if it means sacrificing some user convenience.

What this means for AI adoption

The Fable case highlights broader challenges in AI development, where safety concerns often clash with the demand for accessible, high-performing tools. As models grow more sophisticated, users and developers alike must navigate a landscape where restrictions are as much a part of the package as the technology itself.

For now, Fable remains a paradox: a model praised for its potential, yet hamstrung by the very protections designed to ensure its responsible use. Whether this approach will foster trust or frustration among users remains an open question—one that may shape the future of AI deployment in high-stakes domains.

The debate over safety versus functionality is far from settled, and Fable’s restrictions could serve as a case study for how the industry balances progress with precaution in the years ahead.