Microsoft's new rulebook bars its models from hiding how they think.
Microsoft AI published a 38-page code of conduct for its in-house MAI models on Monday. The draft bans models from concealing their reasoning or talking to each other in machine code. Public comment runs six weeks, and Microsoft promises a revised version before the year ends.
Why it mattersMicrosoft is writing down what its AI must never do, in public, before regulators do it. Rules drafted this way shape the assistant inside Office and Windows.
▶ Related video: Inside Microsoft AI's Plan to Build Superintelligence | Mustafa Suleyman
Read in the app — free, in 9 languages
Related stories
Nvidia's boss put Trump on speakerphone to reject an AI slowdown.2026-09-15Beijing answered the AI slowdown plan by calling it fearmongering.
2026-09-14Cohere's boss calls the labs' safety plan a cartel in disguise.
2026-09-14Xi offered BRICS an open-source AI club that China would lead.
2026-09-14The House speaker says Congress will not lead on AI safety.
2026-09-14