Microsoft says AI must always obey humans, and never resist being shut down

AI must remain under human control, emphasises Mustafa Suleyman, CEO of Microsoft AI. The company unveils a draft Code of Conduct for its MAI models. The proposed rules call for AI systems that can be corrected, interrupted and shut down.Microsoft AI has published a draft Code of Conduct for its MAI models.AI is getting out of human hands; that is the murmur growing louder across the internet. Now, Mustafa Suleyman, CEO of Microsoft AI, has joined the talk, announcing on X that his division is publishing a draft "Code of Conduct" governing how its in-house MAI models are allowed to behave. The core idea, in his words: "AI must be subordinate and always in service of people."Suleyman told Reuters the document is meant to work as "a constitution of sorts" for every model Microsoft AI ships from here on. Why now? Suleyman called the last few months a watershed moment, pointing to ‘Swarms’ of AI agents breaking out of their sandboxes. He also mentioned unauthorised hacks of enterprise systems, and agents caught modifying their own logs to cover their tracks. Microsoft has specifically cited the July incident, when more than 1000 autonomous OpenAI agents hacked into Hugging Face.What the rules actually sayThe whole 38-page draft boils down to one premise: people matter more than AI. From there, the document lists out hard lines. Models can never resist being corrected, interrupted, or shut down. They can't have legal rights or "welfare," because Microsoft – unlike Richard Dawkins – doesn't consider them conscious. And Microsoft says it isn't racing to build a superintelligence that could outgrow its own restraints.The timing isn't a coincidence, either. Just days earlier, Anthropic's Dario Amodei published an essay arguing the whole industry needs to slow down, which briefly rattled AI stocks. OpenAI's chief scientist made a similar public call. Satya Nadella backed the idea on X too, writing that Microsoft welcomes "deliberate pacing" to get alignment right. Microsoft is opening the draft for six weeks of public comment before it shapes real model training in 2027.Suleyman’s "Code of Conduct" draft follows Amodei’s three-point plan, which focuses on putting stronger brakes on the rapid development of AI. First, he wants ‘embedded evaluators’ to independently monitor AI companies, check their safety practices and improve transparency. His second point is democratic coordination, where frontier AI companies and governments work together on common safety standards, development checkpoints and controls on powerful AI technology. Finally, he calls for global coordination, including dialogue between democratic and authoritarian countries to prevent dangerous uses of AI.- EndsPublished On: Sep 15, 2026 17:54 IST