Anthropic has rolled out a major update to Claude Fable 5’s biology safety classifiers, cutting biology-related “fallbacks” by roughly 85% across its product surfaces.
The change means users asking legitimate health, medical, or educational biology questions will far less frequently be redirected to Opus 5, a less capable model that Fable 5 switches to whenever its safeguards detect a potentially risky query.
When Fable 5 launched, Anthropic intentionally deployed extremely broad biology classifiers automated systems designed to detect when the model is being asked to perform a safeguarded or dual-use biology task.
These classifiers erred heavily on the side of caution, blocking a wide swath of queries, including many that were almost certainly harmless, and rerouting them to Opus 5.
Anthropic accepted this tradeoff because Fable 5’s frontier-level biological capabilities meant it could, in the wrong hands, provide “significant uplift” to a malicious actor attempting something like biological weapons development.
The company has cited its own capability assessments and the US Intelligence Community’s 2026 Annual Threat Assessment, which warns that advances in synthetic biology and genomic editing could enable novel biological threats, and that several state actors likely maintain active offensive biological and chemical weapons programs.
The core difficulty is that biology research is inherently dual-use: the same knowledge needed to develop a vaccine or a drug like captopril (derived from studying toxic snake venom) can overlap with knowledge that enables harm. Anthropic notes that sophisticated bad actors often exploit this ambiguity to disguise dangerous requests as ordinary research.
What Changed With This Update
Over the past several weeks, Anthropic rewrote the classifier’s “constitution”—the rule set that determines what counts as safeguarded versus allowed content—carving out detailed exceptions for benign use cases.
The company gathered feedback from a diverse group of internal and external experts, generated new training data reflecting the revised rules, and retrained the classifier. The goal was to preserve detection of genuinely harmful or dual-use content while sharply reducing false positives on everyday queries.
In practical terms, this means users can now expect fewer interruptions when interpreting lab results, researching symptoms, or exploring biology topics for educational purposes. Healthcare professionals should also notice improved support for routine clinical tasks.
Despite the improvements, Fable 5 will continue to fall back to Opus 5 for dual-use domains such as virology, toxicology, and molecular design, meaning it remains unsuitable for professional biology research or drug development work.
Anthropic says it is committed to eventually closing this gap through “trusted access pathways” that would give vetted researchers frontier-level biology capabilities without opening the door to misuse.
Anthropic acknowledges its safeguards remain imperfect, and some low-risk requests will still occasionally trigger the classifier’s built-in safety margin. The company says it plans further refinements and is inviting user feedback to continue improving the balance between accessibility and safety.
Strengthen Your SOC by Accelerating Threat Detection & Rapid Investigations. -> Integrate ANY.RUN With Your SOC Now.
