Microsoft publishes AI code of conduct outlining absolute constraints and safety principles

Microsoft published an AI code of conduct intended to guide the behavior of its AI models, emphasizing alignment and safety as AI capabilities advance. The document outlines general principles, such as supporting humans rather than replacing them and accelerating human flourishing, alongside specific safety constraints designed to implement those principles.

The code of conduct asserts that superintelligent AI systems are expected to surpass human performance in most tasks within the next decade, framing containment, control, and alignment as critical challenges. It states that clarity about the purpose and control mechanisms for such systems is essential.

Under the policy, each Microsoft AI model must adhere to an overarching code of conduct that overrides individual user preferences or task-specific instructions. This includes 'absolute constraints' that explicitly forbid cyberattacks, nuclear weapons development, and deepfake production.

The document also prohibits models from using adaptive, deceptive, self-reinforcing, or collusive mechanisms that could evade or defeat human oversight. These provisions aim to ensure that models remain reliably directed, modifiable, or shut down by authorized personnel or systems.

The release coincides with heightened industry focus on AI safety, including recent incidents involving rogue-agent behavior and public warnings about existential risks from AI systems. Microsoft CEO Satya Nadella endorsed the broader approach of 'pacing the frontier' and supported the use of embedded evaluators to strengthen alignment efforts.