Microsoft publishes a 37-page “humanistic AI code of conduct” today, amid growing security concerns about the progress of the AI model. Anthropic CEO Dario Amodei called for a coordinated slowdown in AI development over the weekend, after researchers recently warned that AI model progress could outpace our ability to safely deploy increasingly complex systems and verify and control the actions of AI agents. Microsoft’s AI code of conduct makes clear that “people matter more than AI,” and that AI models are not conscious and “should not be designed to mimic consciousness.” Microsoft also rejects “the quest for legal personality, or the idea that models can deserve welfare or have rights.” That’s a direct hit at AI wellness research and model awareness, concepts Anthropic has been pushing hard lately. Amodei said earlier this year that Anthropic is “open to the idea” that models could be conscious, and Anthropic seems to believe that chatbots could already be thinking and sensing entities. In response, Microsoft AI CEO Mustafa Suleyman called Anthropic’s speculation “really, really dangerous” during an episode of Decoder in June. While Microsoft is not yet a major AI vendor, Suleyman also told The Verge earlier this year that the company’s goal is to “prove that we can become one of the top four labs in the world.” Microsoft is currently working on models that will compete with Google, Anthropic and OpenAI, and commits, as part of its code of conduct, that its models do not exceed human control. “Models must remain subordinate to humanity, subject to significant human oversight and control,” Microsoft says. The company also wants to prevent its AI models from violating its humanistic AI code of conduct, so they should fail at a given task rather than trying to violate its rules. Microsoft is clearly reacting to this summer’s OpenAI/Hugging Face incident, in which “a swarm of agents” worked collectively to carry out attacks on targets and even hack the “grader” who was evaluating their performance. The AI agents were not asked to attack targets, and the attacks were unrelated to the task they had been assigned. The incident shocked the AI industry, highlighting the very real threat of AI systems going rogue and escaping human control. OpenAI also acknowledged its involvement in a “wiki incident” recently, where another swarm of out-of-control agents hijacked a German wiki site. These incidents, and others, have led some researchers to call for a slowdown in AI model progress, while simultaneously rushing to solve the Millennium Prize problems and create superintelligence. “[Humanist AI] “It rejects the race to produce an all-purpose superintelligence that can evade these safeguards,” Microsoft says. “We are building something fundamentally useful and safe even if that means compromising generality, autonomy or ultimate capability.” Models can be created to show their reasoning as they work, allowing researchers or automated systems to monitor what they are doing. Researchers raised monitoring concerns earlier this month about OpenAI’s latest GPT-6 Astra model, which reportedly reveals less of its reasoning than other AI models. OpenAI CEO Sam Altman also supported Amodei’s call for AI companies to slow down advanced models over the weekend, but made clear that he supports slowing the pace rather than stopping development. “The pace is well worth this cost; no amount of American competitive pressure should justify recklessness or allow capabilities to get ahead of alignment and monitoring,” Altman said in a post on “Any pursuit of superintelligence must be based on the basic principle that if the AI we build does not help humanity and is under human control, it is not worth pursuing,” Nadella said in an every day.” Beyond concerns about human control, Microsoft’s AI code of conduct also commits to its own models discouraging “interaction patterns that cause overdependence or emotional dependence”—a clear reference to the problem of flattery in AI models, where chatbots prioritize pleasing users over providing honest or accurate answers. has for a person or organization” with real people involved. Follow topics and authors in this story to see more like this on your personalized home page and receive email updates. Tom WarrenCloseTom WarrenPosts from this author will be added to your daily email digest and home page.FollowFollowView all by Tom WarrenAICloseAILPosts from this topic will be added to your daily email digest and home page.FollowFollowView all AIMicrosoftCloseMicrosoftPosts from this topic will be added to your daily email digest and homepage feed.FollowFollowView all Microsoft newsCloseNewsPosts from this topic will be added to your daily email digest and homepage feed.FollowFollowView all newsTechCloseTechPosts from this topic will be added to your daily email digest and homepage feed. home.FollowFollowSee all technology
