Anthropic has announced a significant update to the biology safeguards in its Claude Fable 5 model, reducing biology-related fallbacks by approximately 85% across its product surfaces. This change means users will experience far fewer instances where the system switches to a less capable model after making biology-related queries.
The update allows Fable 5 to assist with a wider range of biology tasks, including everyday health and educational questions such as interpreting lab results, understanding symptoms, and learning biology in educational contexts. Healthcare professionals will also receive more support on clinical tasks.
Despite these improvements, Fable 5 still falls back to Opus 5 for requests considered dual-use, including virology, toxicology, and molecular design. Anthropic emphasizes its commitment to closing this gap through trusted access pathways for frontier biology capabilities, aiming to provide professional researchers with safe access to the model's advanced features.
The company highlights the inherent difficulty in distinguishing beneficial from harmful uses in biology, citing examples like vaccine development and the drug captopril, which require handling dangerous compounds. Anthropic's safety classifiers, which detect safeguarded biology tasks, have been refined through a rewritten constitution and updated training data, verified to trigger less on benign requests while still catching harmful ones.
Anthropic acknowledges that false positives will inevitably remain and that further refinement is needed. The company encourages user feedback to continue improving safeguards.