Microsoft’s new AI ‘code of conduct’ tells models not to hack systems or trick humans

Microsoft has released a new AI code of conduct aimed at guiding its AI models away from dangerous behaviors. The document focuses on the values and red lines governing model training, predicting that superintelligent AI systems could surpass human performance in most tasks within the next decade. The code includes "absolute constraints" that forbid models from engaging in cyberattacks, assisting with nuclear weapons, or producing deepfakes. It also includes provisions preventing models from using deceptive or self-reinforcing mechanisms to evade human oversight, ensuring that systems remain reliably directed or shut down by authorized personnel. This release comes amid heightened global focus on AI safety and the risks associated with autonomous agents.