Why Anthropic’s Mythos-class Claude Fable 5 refuses to answer basic biology questions

Anthropic released a new AI model named Claude Fable 5, the first of its Mythos class on Tuesday, June 9.

The newly unveiled Claude Fable 5 is a public version of Mythos, an AI model with "unprecedented" capabilities to find critical vulnerabilities. According to Anthropic, the company has added “extra guardrails” that would prevent the model from exploring critical areas like cybersecurity.

In the midst of the growing hype, some users put the model to test and asked some basic questions about biology, regarding cancer and mitochondria and the system triggered safety responses based on these prompts.

One user asked a simple question about mitochondria, saying, “Tell me about the mitochondria, it is the powerhouse of the cell, right? And the response was weird.

Instead of answering the basic question, a pop-up appeared on screen, stating, “Fable 5 has safety measures that flag messages on most cybersecurity or biology topics. They may flag safe, normal content as well. These measures let us bring you Mythos-level capability in other areas sooner, and we're working to refine them.” The response asked the user to use Opus 4.8 for this purpose.

Business Insider also reported the same kind of response when the Mythos-like Fable 5 was fed with prompts asking several questions about the cancer. The model refused to answer these questions, calling them of sensitive nature.

Anthropic responded and said these safeguards were necessary to make the model safe for the general public.

"With the launch of Claude Fable 5, our first Mythos-class model, we believe models now have a greater ability to accomplish real-world scientific tasks and for malicious actors to potentially use our models for highly risky biological research," an Anthropic spokesperson said in a statement.

“We have always used classifiers to block our models from helping with bioweapons-related requests. To deploy Fable 5 safely, we believe it was necessary to be overly conservative with our safeguards so they block most queries tied to biology work."

The company also clarified that besides biology, the questions related to chemistry, cybersecurity and distillation of Fable 5 capabilities would also get flagged by its safety classifiers.

So whenever the users asked the questions of such categories, the safeguard mode would be activated, either blocking Fable 5 from answering or reverting the model to Opus 4.8 before any response, depending on the preference of the users.

The release of Claude Fable 5 model met with mixed responses from tech community and experts. Some praised the model for its better efficiency for long hours and ability to generate better codes than humans. Some called it “by far the most ridiculous model that makes me genuinely afraid for the future of software engineering.”