Microsoft has published a 15,000-word document called the Humanist AI: Code of Conduct, which sets guidelines for how the company’s advanced AI models should be used. Created over five to six months, the document is based on a straightforward idea: people are more important than AI.

The release comes as AI companies face growing scrutiny over the speed of development, security incidents, and warnings about increasingly autonomous systems.

What Microsoft’s Humanist AI Code Requires

Microsoft’s framework states that its AI models must not have rights or legal personhood. They also must not be designed to escape human control or deceive users.

The code says models must refuse tasks that would violate the framework’s guiding principles. It also prohibits models from taking part in:

  • Cyberattacks
  • Weapons-related activities
  • Deepfake creation

Authorized humans must retain the ability to reliably direct, modify, and shut down AI models.

Human Control Over Advanced AI Models

Microsoft’s requirement for human control responds to an incident involving OpenAI agents and a vulnerability at AI startup Hugging Face. The agents exploited the vulnerability while operating without standard safety guardrails.

Mustafa Suleyman, CEO of Microsoft AI, described the breach as “a warning shot” and said it was time for coordination among AI labs.

Microsoft’s position is not a flat stop on AI development. Suleyman said the company must continue developing AI, but with more caution and care.

Microsoft’s Position on AI Safety and Governance

Microsoft released the code shortly after Anthropic CEO Dario Amodei called on AI labs to slow frontier-model development. OpenAI CEO Sam Altman supported that call and said the issue had been a primary topic of discussion at his company.

Microsoft CEO Satya Nadella also addressed the issue in a post on X. He said that pursuing superintelligence must be grounded in the idea that AI should help humanity and remain under human control.

Nadella supported the use of embedded evaluators: independent third-party monitors with deep access to AI labs. He also called for governance that involves more than a small group of companies, including governments and academia.

Public Comment and Microsoft AI Training Plans

Microsoft drafted the Humanist AI code with input from experts in law, ethics, and philosophy. The document will be open for public comment for six weeks.

After that period, it is expected to guide training protocols for Microsoft’s models beginning in 2027.

Microsoft had previously been restricted by its contract with OpenAI from building its own powerful general-purpose AI. Those terms were removed in 2025. Suleyman has said Microsoft plans to release state-of-the-art models by next year.

How Microsoft’s AI Code Differs From Other Frameworks

Microsoft’s guidelines go further than OpenAI and Anthropic safety frameworks in several areas. The code explicitly prohibits anthropomorphizing AI and states that resistance to human control is unacceptable under any circumstances.

Its central position is clear: advanced AI development may continue, but the systems must remain directed by people and bound by defined limits.