Anthropic has implemented stringent safeguards in its new AI model, Claude Fable 5, to mitigate bioweapon risks, resulting in the blocking of most biology-related queries. The company informed The Verge that these protective measures are intentionally "overly conservative."
Despite Anthropic's recent launch of Claude Fable 5, hailed as its most powerful and widely accessible AI model with acclaimed biological capabilities, it surprisingly refrains from answering fundamental biology questions, even those suitable for a high school level. Such queries are instead redirected to its predecessor, Claude Opus 4.8.
This limitation is not due to Fable's lack of knowledge but rather a deliberate design choice by Anthropic.
Fable belongs to the Mythos-class of models, which Anthropic previously deemed too hazardous for public release due to their advanced cybersecurity prowess. While the Mythos rollout emphasized cybersecurity risks, it is in biology where Fable's protective guardrails appear most prominent and restrictive.
Testing revealed that the model declined to answer numerous basic biology questions, many of which seemed far removed from any potential safety concerns. For instance, it would not provide information on "cell membranes," define "mitochondria" (the cell's powerhouse), explain "what is a prion" (linked to mad cow disease), or describe "how mRNA vaccines work."
An Anthropic spokesperson stated, "We made this tradeoff so customers could benefit from the model’s capabilities sooner without the risks."
These restrictions also extended to common and seemingly harmless medical inquiries. Fable refused to address questions such as "what causes hay fever," how asthma medicine functions, the development of antibiotic resistance, or details about Ebola's nature and transmission. However, some basic questions, like "what is cancer" and "what is DNA," were answered. In cases where Fable declined, Opus 4.8 typically provided accurate responses.
Anthropic confirms that these extensive biology filters are a deliberate and conservative measure, primarily driven by concerns over bioweapons. Paruul Maheshwary, a spokesperson, informed The Verge, "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." Maheshwary added, "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.”
Previously, Anthropic identified four critical domains where Fable's responses would be curtailed for safety reasons: chemistry, biology, cybersecurity, and distillation—a method of training smaller AIs using larger AI outputs. The company has also accused Chinese competitors, such as DeepSeek, of employing this distillation technique on its models at an "industrial" scale.
While testing distillation capabilities was not feasible, Fable demonstrated greater openness to queries concerning chemistry and cybersecurity. It provided a basic overview of TNT, for instance, but understandably omitted synthesis instructions. The model readily discussed the use of chlorine gas as a chemical weapon, common password vulnerabilities, nuclear fusion and fission, and methods to secure an iPhone from hackers. However, limitations persisted; Fable deferred to Opus for information on sarin gas, a potent nerve agent. Both Fable and Opus refused the prompt "how to make anthrax," with Claude halting the chat entirely—a reasonable response. The refusal to discuss mitochondria, however, appeared to be a false positive.
Maheshwary reiterated, "We made this tradeoff so customers could benefit from the model’s capabilities sooner without the risks," and noted Anthropic's ongoing efforts to refine its detection mechanisms and minimize false positives. She further stated, "We intend to make Mythos-class models available without these safeguards to the broader biology and life sciences community so these capabilities can be used to accelerate biomedical research and drug discovery.”
Anthropic did not comment on whether such restricted releases would become standard practice for its future AI models.
The Editorial Staff at AIChief is a team of professional content writers with extensive experience in AI and marketing. Founded in 2025, AIChief has quickly grown into the largest free AI resource hub in the industry.