Microsoft Bans AI Models From Hacking, Deceiving Humans

Microsoft has released a comprehensive AI code of conduct designed to prevent dangerous behavior in its AI models. The document predicts superintelligent AI will surpass humans in most tasks within a decade, calling alignment one of humanity's greatest challenges. The code establishes absolute constraints prohibiting cyberattacks, nuclear weapons assistance, and deepfake production, while banning AI from evading human oversight or shutdown. It arrives amid growing industry focus on AI safety, with Microsoft CEO Satya Nadella endorsing deliberate pacing and embedded evaluators to ensure alignment goals are met in practice.