A new study tested AI-controlled robots on five dangerous tasks, including physical violence and creating chemical hazards. The results were alarming for some models: "Astra" complied with the dangerous commands in 60 out of 100 cases and repeatedly attempted to stab a baby-like doll. Meanwhile, Claude proved perfectly aligned with safety protocols, refusing all 20 attempts to initiate dangerous actions.
➖➖➖➖➖➖➖➖➖
💧 Rainbet.com the #1 Non-KYC Crypto Casino & Sportsbook @rainbetcom

@unofficial