aiqbench analysis

Anthropic has refined its biology safety classifiers in Claude Fable 5 to reduce false positives by approximately 85%, allowing legitimate medical and educational queries to bypass unnecessary model fallbacks while maintaining safeguards against dual-use risks. This update reflects a strategic recalibration of the constitutional rules governing frontier AI capabilities, balancing security concerns regarding biological threats with the need for accessible scientific information. The change addresses early over-caution that inadvertently blocked harmless research by incorporating more granular exceptions based on threat assessments and community feedback.