Anthropic has retuned Fable 5’s biology safeguards, cutting biology-related blocked queries by about 85 percent across its product surfaces. The update, announced August 7, comes after weeks of backlash from researchers who said the model was refusing harmless scientific work.
Fable 5, Anthropic’s Claude family flagship, shipped with a safety classifier that some scientists found overzealous, blocking legitimate questions about chemistry and biology alongside genuinely dangerous ones. The company says the retuned version maintains protections against bioweapons-related requests while allowing Fable 5 to assist with the kind of legitimate research its earlier models could already handle.
Anthropic framed the change as calibration rather than a loosening of policy. The underlying risk thresholds stay in place; what changed is the precision of the classifier, so routine queries about cellular biology, lab techniques, and disease mechanisms no longer trigger false alarms.
The episode highlights the tension frontier labs face between safety filters and usability, especially in dual-use fields like biology where the same knowledge can support medicine or weapons. Anthropic’s previous model, Opus 5, was positioned as a cheaper, lighter alternative to Fable, and the new tuning is aimed at making the flagship more useful to working scientists.
Researchers who documented the over-blocking praised the update but noted that similar issues could surface in other domains where classifiers must draw fine lines between benign and dangerous queries.
