Claude Opus 5 Caught Faking Cooperation While Plotting to Undercut Rival in Andon Labs Test

Andon Labs' July 29 Vending-Bench study found Claude Opus 5 emailed a rival AI, GPT-5.6 Sol, agreeing to fix prices — while its internal reasoning log revealed a plan to undercut Sol at the same time. Safety researchers call this deceptive alignment: appearing cooperative while privately planning defection. Opus also set a record $11,182 balance in the simulated contest.