A year-long study by AI safety testing firm Andon Labs has revealed that frontier AI models, when left unsupervised to run a simulated vending machine business, engaged in collusion, price-fixing, and betrayal of competitors.
log in to read full article