Anthropic’s Claude Fable 5 won’t answer basic biology questions, and that’s by design

Anthropic launched Claude Fable 5 on June 9, 2026, calling it the most powerful AI model the company has ever made widely available. It praised the system’s skills in biology, among other domains. There’s just one problem: it won’t actually answer basic biology questions.

The kind of questions a high schooler could handle get punted to Claude Opus 4.8, the former flagship model. Not because Fable 5 doesn’t know the answers. Because Anthropic won’t let it.

The safety reroute, explained

Claude Fable 5 is classified under Anthropic’s “Mythos-class” category and shares its underlying weights with Claude Mythos 5, the company’s most advanced system. Mythos 5 itself is restricted to trusted partners through a program called Project Glasswing. Fable 5 is the version the rest of us get to use.

Fable 5 comes loaded with safety classifiers that detect queries in sensitive domains, specifically biology, chemistry, and cybersecurity, and automatically route them to the older Opus 4.8 model. In English: ask it something Anthropic deems potentially dangerous, and you’re quietly talking to last year’s AI instead.

Over 95% of user sessions reportedly don’t trigger the reroute at all. Most people asking Claude about project management or writing code will never notice the guardrails. But for researchers, students, and professionals working in the sciences, the limitation is immediately apparent, and immediately frustrating.

What it costs and where to find it

Fable 5 is available through the Claude API and Amazon Bedrock, priced at $10 per million input tokens and $50 per million output tokens. Anthropic is offering introductory pricing incentives that run until June 22, 2026, though the company hasn’t specified the discount structure publicly.

The model is purpose-built for complex reasoning, long-term task execution, and knowledge-based work.

Why crypto should pay attention

Anthropic has no direct ties to any crypto protocols or tokens. The company hasn’t launched a blockchain product, issued a token, or partnered with a DeFi protocol. But Claude’s various iterations have quietly become tools of the trade in digital asset security.

Previous versions of Claude have been used for smart contract vulnerability detection, blockchain security auditing, and crypto trading signal analysis. In simulation exercises, earlier Claude models identified potential exploits valued at up to $4.6 million.

The safety reroute in Fable 5 creates an interesting wrinkle for these use cases. Cybersecurity is one of the domains where sensitive queries get shunted to the older model. If you’re a security researcher trying to use Fable 5 to probe a smart contract for vulnerabilities, you might find yourself working with Opus 4.8’s reasoning capabilities instead of the frontier model you’re paying for.

Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.