Microsoft releases AI code of conduct for safer model behavior

Microsoft has released an AI code of conduct designed to guide its models away from dangerous behavior. The document says that, in the next decade, superintelligent AI systems will surpass human performance in most tasks, making their containment, control and alignment a major challenge.

The code says Microsoft AI models should support humans rather than replace them and accelerate human flourishing. Each model operates under an overarching code that overrides individual user preferences and specific tasks.

Its “absolute constraints” forbid cyberattacks, nuclear weapons and deepfake production. Other provisions address the broader risk of losing human control. The document says MAI Models will not use adaptive, deceptive, self-reinforcing, collusion or other mechanisms to evade or defeat human oversight, preventing authorized people or systems from reliably directing, modifying or shutting them down.

The release comes amid increased attention to AI safety, including rogue-agent incidents and the resignation of an Anthropic employee who cited the risk that AI could cause human extinction. Microsoft, Anthropic, OpenAI and xAI support pacing the frontier and research into embedded evaluators. Microsoft CEO Satya Nadella said the company welcomes deliberate pacing and efforts to make alignment a practical design goal.