Written by Admin Alex · Fact-Checked by M.Ali · Info Verified September 2026
We review and update this article regularly as new information becomes available.
TL;DR: Microsoft published a code of conduct for its AI models that bans cyberattacks, nuclear weapons development, deepfakes, and any attempt by a model to deceive the humans overseeing it, framed as an “absolute constraint” that overrides any individual user request. CEO Satya Nadella says it reflects the deliberate pacing needed to get AI alignment right.
Don’t hack systems. Don’t help build nuclear weapons. Don’t create deepfakes. Don’t lie to the humans supposed to be supervising you. Those are, in plain terms, some of the “absolute constraints” Microsoft just wrote into a new code of conduct governing how its AI models are supposed to behave, no matter what a user asks them to do.
The document sits on two levels. At the top are broad principles: AI should support humans rather than replace them, and it should advance human flourishing rather than just chase capability for its own sake. Underneath that sit specific, hard lines the models aren’t supposed to cross, and Microsoft is explicit that these constraints sit above individual tasks or user preferences. A user asking a model to help with something on the banned list doesn’t get to override the rule by asking nicely or framing it as research.
Why Microsoft is doing this now
The company’s own framing is stark. Microsoft predicts superintelligent AI, systems that meaningfully surpass human capability, could arrive within a decade, and it’s calling alignment “one of the greatest challenges humanity has ever faced.” That’s not typical corporate language for a policy document, and it signals the release wasn’t written purely for optics.
The timing lines up with a rougher few months for AI safety optimism generally. There have been incidents involving AI agents behaving in unintended ways, and a researcher at Anthropic publicly resigned while warning about extinction-level risk from advanced systems. Satya Nadella’s public comments alongside the release lean into that mood, calling for the “research, focus, and deliberate pacing needed to get alignment right” and floating ideas like embedded evaluators, essentially built-in watchdogs, as a practical oversight mechanism.
Where this fits with everyone else
Microsoft isn’t operating in a vacuum here. Anthropic, OpenAI, and xAI have all separately signaled support for some version of deliberate pacing in frontier AI development, most visibly through Dario Amodei’s recent public call to slow capability gains. What makes Microsoft’s move distinct is that it’s not a strategic statement about industry pace, it’s an implementation-level document, specific rules baked into how the models are actually supposed to behave in the moment, not just a philosophy about how fast to build them.
Bottom Line
A code of conduct is only as good as the enforcement behind it, and Microsoft hasn’t detailed exactly how these constraints get tested or verified in practice. Still, writing “don’t help build nuclear weapons” into a formal AI policy document, and meaning it as a genuine technical constraint rather than a throwaway line, tells you something about where the industry’s internal conversation has actually moved. Six months ago, this read like science fiction caution. Now it’s a released policy.
