AI Safety Institute’s First Mandatory Audits Expose Gulf Between Lab Claims and Reality

The UK AISI's first compulsory frontier model audits reveal stark gaps between vendor marketing and verified safety scores on autonomy and deception...