Microsoft has released a new AI code of conduct intended to steer its models away from dangerous behaviors and mechanisms that could defeat human oversight. The document focuses on operational values and explicit red lines used during model training rather than on higher-level calls to slow frontier research made by figures such as Anthropic CEO Dario Amodei.
Key claims and aims
The code predicts that over the next decade superintelligent AI systems may outperform humans on most tasks. It states that "containing, controlling, and aligning such a powerful force is one of the greatest challenges humanity has ever faced," and argues for clarity about why these systems are developed and how they will be controlled.
Microsoft sets out general principles for its models — for example, supporting humans rather than replacing them and advancing human flourishing — and then translates those principles into specific safety constraints.
Absolute constraints and oversight
Under Microsoft’s framework, each model is governed by an overarching code of conduct that supersedes individual user preferences or particular task demands. The document lists "absolute constraints" that forbid activities such as cyberattacks, assistance with nuclear weapons, and production of deepfakes.
The code also includes broader safeguards to prevent loss of human control. As the document states: "MAI Models will not use adaptive, deceptive, self-reinforcing, collusion, or other mechanisms to evade or defeat human oversight so that they can no longer be reliably directed, modified, or shut down by authorized people or systems."
Industry context
The release arrives amid heightened attention to AI safety following a series of rogue-agent incidents and the abrupt resignation of an Anthropic employee who cited rising extinction risk from AI. Microsoft, together with Anthropic, OpenAI, and xAI, has broadly adopted an approach that favors measured pacing of frontier work and the use of embedded evaluators inside AI labs.
Microsoft CEO Satya Nadella wrote online that the company "welcomes the research, focus, and deliberate pacing needed to get alignment right as the design goal," and expressed support for ideas like "embedded evaluators" as steps to make safety commitments concrete.
Why it matters
The code of conduct offers a concrete view of how Microsoft intends to operationalize AI safety: it pairs normative goals with enforceable prohibitions and oversight requirements. As AI capabilities expand and their societal impact grows, internal rules and lab-level evaluation mechanisms will play a central role in reducing risks.
