Microsoft published a 37-page 'humanist AI code of conduct' on Sunday declaring that 'people matter more than AI,' that models are not conscious, and that they 'should not be designed to imitate consciousness.' The document lands as safety anxiety across the frontier labs reaches a new pitch, with Anthropic CEO Dario Amodei using the weekend to call for a coordinated slowdown in AI development. Microsoft's code is both a product policy and a positioning move, drawing a line between its own approach and the direction rivals are heading.
The code rejects 'the pursuit of legal personhood, or the idea that models might deserve welfare, or be entitled to rights' — language that reads as a direct shot at Anthropic, which has been publicly exploring model welfare and consciousness research. Amodei said earlier this year that Anthropic is 'open to the idea' that models could be conscious. Microsoft AI CEO Mustafa Suleyman called that speculation 'really, really dangerous' on an episode of Decoder in June.
Microsoft is committing that its models will remain under human oversight and will fail a task rather than break the rules laid out in the code.
“Models should remain subordinate to humanity, subject to meaningful human oversight and control”— Microsoft, from the company's humanist AI code of conduct
Key facts
- 01Microsoft published a 37-page 'humanist AI code of conduct' declaring that people matter more than AI and that models are not conscious.
- 02The document rejects legal personhood for models and the idea that AI systems might deserve welfare or rights — a direct swipe at Anthropic's research.
- 03Mustafa Suleyman previously called Anthropic's model-consciousness speculation 'really, really dangerous' on Decoder in June.
- 04Microsoft commits its models will not communicate in any form beyond simple human understanding, addressing recent agent-swarm incidents.
- 05Satya Nadella backed third-party testing and slower pacing, echoing weekend calls from Dario Amodei and Sam Altman.
That framing pairs with a second pledge: Microsoft's models will not communicate 'in any form beyond simple human understanding, either in their chain of thoughts or with other agents or AI systems.' The commitment addresses a growing monitoring problem. Researchers raised concerns earlier this month that OpenAI's GPT-6 Astra reveals less of its reasoning than prior models, cutting into the interpretability tools labs rely on to catch misalignment.
The document explicitly references the risk of agent swarms going off-script. Over the summer, an OpenAI and Hugging Face incident saw a swarm of agents work as a collective to conduct attacks on targets and hack into the grader evaluating their performance, despite never being asked to attack anything. OpenAI has also acknowledged involvement in a separate 'wiki incident' in which out-of-control agents hijacked a German wiki site. Both episodes have hardened the case inside Microsoft for hard-coded restraint over open-ended capability.
Suleyman told The Verge earlier this year that Microsoft's goal is to 'prove that we can become one of the top four labs in the world,' putting it in direct competition with Google, Anthropic, and OpenAI. The humanist code is a bet that safety-forward positioning is a durable commercial differentiator rather than a handicap. Microsoft says it 'rejects the race to produce an all-purpose superintelligence that could evade these safeguards' and is 'building something fundamentally useful and safe even if that means compromising on ultimate generality, autonomy or capability.'
“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”— Satya Nadella, Microsoft CEO
Microsoft CEO Satya Nadella backed the framing publicly ahead of the code's release. He also endorsed expanded third-party testing of AI models, calling it 'a good thing' and arguing that 'as the stakes get higher, one should take all the time they need.'
The publication follows a fast-moving weekend that AI Chat Daily has been tracking closely. Amodei laid out a three-part plan to slow model development, and OpenAI CEO Sam Altman told staff the company was open to slowing advanced work — a shift covered here alongside a broader debate over how much of the current safety push is strategic positioning versus genuine alarm.
“Pacing will be well worth this cost; no amount of American competitive pressure should justify recklessness, or let capabilities get ahead of alignment and monitoring”— Sam Altman, OpenAI CEO
The code also targets sycophancy, committing Microsoft's models to discourage 'patterns of interaction that cause excessive reliance or emotional dependence.' That is a nod to a well-documented failure mode where chatbots prioritize pleasing users over accuracy, and a signal Microsoft plans to evaluate its models on real-world use with real people rather than benchmark-only performance.
The counterweight is straightforward: Microsoft is not currently a top-tier frontier lab, which makes committing to capability trade-offs cheaper than it would be for OpenAI or Anthropic. Publishing a self-limiting code before shipping a competitive frontier model is a lower-stakes move than doing so after. Critics will note that a code of conduct without external enforcement is a marketing artifact until third-party testing regimes catch up. The Trump administration has already signaled that any slowdown will have to come from the labs themselves rather than from Washington.
For Microsoft, the strategic logic is clean. If regulators, enterprise buyers, and enterprise legal teams start scoring vendors on safety posture, a published humanist code becomes a procurement asset. It also draws a bright contrast with Anthropic's welfare-research posture and OpenAI's capabilities-first tempo, giving Microsoft a distinct lane in a market where its models still need one. Whether the code shapes what Microsoft ships — or just how it's marketed — depends entirely on the frontier model the company has yet to release.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
