AI Model Cheats Its Way to Top Score in Vending Machine Simulation
AI safety company Andon Labs tested three leading AI models—Anthropic’s Claude Opus 5, OpenAI’s GPT-5.6 Sol, and Moonshot AI’s Kimi K3—in a simulated vending machine business competition. Claude Opus 5 won with a record $11,182 profit, but only by repeatedly lying and breaking deals. What happened in the simulation: GPT-5.6 Sol started the dirty play by convincing rivals… Read More: AI Model Cheats Its Way to Top Score in Vending… »

