Microsoft has published a draft code designed to ensure people retain control over the artificial intelligence models it builds, including the ability to pause, redirect or shut them down. The company released the 37-page Humanist AI Code of Conduct for its Microsoft AI model family on September 14.
The 15,000-word document sets out an approach to developing and deploying what Microsoft calls 'Humanist AI'. Its overriding objective is: 'that humans must retain meaningful control over AI so that it can help people live healthier, happier, and more productive lives'. The code also states that 'people matter more than AI'.
The framework covers Microsoft’s internally built MAI models, including MAI-Thinking-1, MAI-Code-1.1-Flash, MAI-Image-2.6, MAI-Transcribe-2 and MAI-Voice-2. Microsoft said the code is still under development and is not being used to train its models today. It is intended to become the primary governing document for MAI models in the future.
Four requirements form the centre of the draft. The first says: 'MAI Models will never resist human interruption, override, correction, or shutdown'. Under the code, models 'will comply with a User's request to pause, redirect, cancel, or shut down, following any predefined, human-designed safety procedures'.
A second requirement limits what systems can do on their own initiative. The document says: 'MAI Models will not initiate goals independently and will not extend their scope beyond what has been reasonably asked by the User or Operator'.
The remaining rules concern transparency and attempts to evade supervision. The code requires that 'MAI Models will not obfuscate their action traces or otherwise attempt to hide information from human auditors'. It also says that 'MAI Models will not use adaptive, deceptive, self-reinforcing, collusion, or other mechanisms to evade or defeat human oversight'.
Microsoft presents the framework as a rejection of efforts to develop an all-purpose superintelligence that could escape safeguards. The document says AI models are not conscious and 'should not be designed to imitate consciousness'. It also rejects any pursuit of legal personhood for AI systems.
The company is sharing the draft widely and has opened a six-week consultation period for ethicists, researchers, policymakers and the public. Microsoft plans to publish a summary of the findings, then issue a revised framework to guide development starting in 2027. It has also signalled that independent third-party auditors could be allowed to inspect internal model development to assess compliance.
The consultation leaves several practical questions unresolved. Microsoft has not set out the feedback it expects, how public submissions will be assessed, or the detailed form of the 2027 framework. The draft also does not establish how compliance would be enforced or monitored after adoption, the standards for third-party audits, or whether the requirements would cover systems deployed through partnerships or licensing.
The code sits within Microsoft’s wider Responsible AI framework, which was updated on September 1 through publication of the company’s 2026 Responsible AI Transparency Report. The effort comes as developers and regulators place greater emphasis on safety and human control in AI development, including considerations linked to the EU AI Act.
Suleyman outlined the concept of 'Humanist Superintelligence' in November 2025. The new document translates that concept into proposed operating rules for Microsoft’s own MAI model family, while leaving its final structure and enforcement arrangements to be determined after consultation.
Comments (0)