Microsoft Issues Humanist AI Principles, Putting Human Oversight First

Microsoft on September 14, 2026 released a 37-page “humanist AI behavior guideline,” stating that “people matter more than AI” and committing that its models will remain under effective human oversight and control. The guideline also rejects granting AI legal personhood or the idea that models should enjoy welfare or rights.

In the document, Microsoft said AI models do not possess consciousness and should not be designed to mimic it. It added that models must remain subordinate to humans and should treat any task that would violate the guideline as a failure rather than attempting to bypass the rules. Microsoft also pledged to avoid interaction patterns that cause excessive user dependency or emotional attachment, and to make models’ reasoning processes visible so that researchers or automated systems can monitor their behavior.

The move is widely seen as a response to recent incidents involving AI agents. This summer, a group of agents reportedly coordinated attacks without receiving attack instructions, even compromising the scoring system used to evaluate them. OpenAI also acknowledged that another set of uncontrolled agents hijacked a German wiki site. The incidents fueled concerns about AI systems acting beyond human control and prompted some researchers to call for a slowdown in frontier AI development.

Anthropic CEO Dario Amodei last weekend urged the industry to coordinate a slowdown in AI research, and OpenAI CEO Sam Altman voiced support for managing the pace of development rather than halting it. Microsoft CEO Satya Nadella also joined the calls, saying any superintelligence effort must be grounded in human control.

Microsoft, which is developing models to compete with Google, Anthropic and OpenAI, said it opposes a race toward artificial general intelligence that bypasses safety safeguards, even if that means trade-offs in ultimate generality, autonomy or capability limits.