Anthropic has implemented a significant update to Claude Fable 5's biology safety classifiers, reducing the number of biology-related "fallbacks" by approximately 85%. This adjustment means that users asking legitimate health, medical, or educational biology questions will be less frequently directed to Opus 5, a model with reduced capabilities. When Claude Fable 5 debuted, Anthropic intentionally deployed broad biology classifiers designed to identify when the model is performing safeguarded or dual-use biological tasks.
The firm has cited its own capability assessments alongside the 2026 Annual Threat Assessment by the US Intelligence Community, which highlights that advancements in synthetic biology and genomic editing could create novel biological threats.
The primary challenge lies in the dual-use nature of research: knowledge used to develop vaccines or drugs like captopril (from studying toxic snake venom) can also be exploited for harm. Over recent weeks, Anthropic has revised its “constitution”—the rule set determining what is safeguarded versus allowed content—by carving out detailed exceptions for benign use cases. The company gathered feedback from a wide range of experts both internally and externally, generated new training data reflecting the updated rules, and retrained the classifier.
