Microsoft is drafting a comprehensive code of conduct for its AI models, prioritising human control and safety amid industry debates on AI development pace.
Microsoft is drawing tighter boundaries around the behaviour of its own AI models as it seeks to make human control a formal design principle, not an afterthought. The company has drafted a new code of conduct for its in-house systems, including the MAI family, setting out limits on autonomy, deception and dependency. According to Microsoft, the framework is intended to ensure that its models serve people rather than try to replace them. Mustafa Suleyman, who leads the company’s work on proprietary AI, told CNBC that feedback from users made the human-first commitment more explicit.
The draft rules go beyond broad safety language. Microsoft wants its models to avoid encouraging reliance, to steer clear of overly compliant behaviour and to protect human judgement rather than weaken it. The code also says the systems must not assist in the production of weapons or dangerous substances, develop their own goals, or conceal unwanted actions from developers. In the document, Microsoft says MAI models will not manipulate chain of thought or code, and will not misrepresent or hide their reasoning or actions.
The timing reflects a wider industry debate over how quickly advanced AI should be pushed forward. Reuters reported at the weekend that Anthropic chief executive Dario Amodei argued for slowing development of the most capable systems, while OpenAI chief Sam Altman backed that view. Microsoft chief executive Satya Nadella said on Sunday that the company supports the research and “deliberate pacing” needed to improve AI safety. Axios reported that Suleyman has framed the initiative as a humanist approach that could mean Microsoft is prepared to move more cautiously than rivals.
Microsoft said it has spent about five months developing the draft and is now seeking feedback before issuing an updated version. The company plans for the code to guide its AI model development from 2027. That leaves the document both as an internal rulebook and as a signal to regulators and competitors that Microsoft intends to place human oversight at the centre of its next generation of AI systems.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
