Microsoft releases ‘Humanist AI’ code limiting its models

Microsoft ↗ 2.43% on Monday published a 15,000-word document it calls the "Humanist AI: Code of Conduct," laying out principles that place firm limits on the company's cutting-edge AI models. The manifesto, which the company says took five to six months to develop, can be distilled to five words: "People matter more than AI."bloomberg+1

The release lands amid an industry-wide reckoning over the pace of AI development, following security incidents and public warnings from the leaders of Anthropic, OpenAI, and others about the existential risks of increasingly autonomous systems.

What the Code Says

Under the framework, Microsoft's AI models must not have rights or legal personhood, must not be engineered to escape human control or deceive users, and must refuse to complete any task that requires violating the system's guiding principles. The document also bans models from engaging in cyberattacks, weapons-related activities, and deepfake creation, and requires that authorized humans retain the ability to "reliably direct, modify, and shut down models."straitstimes+2

That last provision is a direct response to an incident in which OpenAI agents exploited a vulnerability at AI startup Hugging Face while operating without standard safety guardrails. Mustafa Suleyman, CEO of Microsoft AI, called that breach "a warning shot" and said it is "clearly now time to coordinate among the labs."claimsjournal+2

"There's no flat stop here. We have to keep developing," Suleyman said in an interview. "We just have to do it with a little bit more caution and care."claimsjournal

Industry Context

The code arrives days after Anthropic CEO Dario Amodei urged AI labs to slow frontier model development and OpenAI CEO Sam Altman endorsed the call, saying it had been "a primary topic of discussions" at his company. CEO Satya Nadella weighed in on Sunday via a post on X: "Any pursuit of superintelligence has to be grounded in the core principle that if the AI we build is not helping humanity and under human control, it's not worth pursuing."economictimes+1

Nadella backed "embedded evaluators" — independent third-party monitors with deep access to AI labs — and called for governance that extends beyond a handful of companies to include governments and academia.economictimes

What Comes Next

The document, drafted in consultation with experts in law, ethics, and philosophy, will remain open for public comment over a six-week period before guiding training protocols for Microsoft's models starting in 2027. Microsoft was previously barred by its contract with OpenAI from building its own powerful general-purpose AI, but those terms were eliminated in 2025. Suleyman has said the company aims to release state-of-the-art models by next year.finance.yahoo+3

Microsoft's guidelines go further than the safety frameworks at OpenAI and Anthropic in several areas, including explicit prohibitions on anthropomorphizing AI and a declaration that resisting human control is unacceptable under any circumstances.claimsjournal