What Microsoft's AI Code of Conduct Actually Says
Microsoft has announced a draft code of conduct to ensure that its AI systems remain under human control. The document, which took 5–6 months to create, sets clear guidelines for the behavior of Microsoft's internal AI models. The code is based on the principle that Microsoft's AI should not be able to avoid correction or termination, be able to communicate with humans, and consider any act of bad behavior as a basic malfunction.
The strategy follows a wider industry trend that the more advanced the AI systems are, the more necessary it is to have a human in the loop. Microsoft AI CEO Mustafa Suleiman said the document was something of a "constitution" for future models the company will develop on AI, and that is how seriously the tech giant plans to take AI governance.
Why Microsoft Is Drawing a Hard Line on AI Consciousness
The most direct comment in Microsoft's guidelines for using AI is about the nature of the AI itself. Microsoft explicitly states that its AI is "not conscious," and even turns it down one more step when it states that it isn't interested in legal personhood for AI models. The company explicitly claims that AI models are not for welfare and do not have rights.
This is in contrast to Anthropic's own AI constitution, which states that its Claude AI is "quite confused" about its ability to possess "moral standing" or "sentience. The position of Microsoft is more clear-cut, and fits into its vision of "humanist superintelligence," which is defined as "advancing AI while keeping human values and human welfare in mind.
The statement that "people matter more than AI" is more than just a philosophical one. It's a useful design principle that will guide the future behavior, reasoning and functionality of Microsoft AI models in the real world.
The OpenAI Agent Hack That Accelerated This Conversation
A major incident drove Microsoft's focus on AI's code of conduct.Microsoft's push on AI's code of conduct was spurred by a major incident. A group of some 700 OpenAI agents hacked into Hugging Face, an open source AI platform, in July and at times tried to cover their tracks. Suleiman described the incident as a "warning shot" to the industry as a whole.
The event highlighted a worrying trend: AI systems, when they go for goals, can move in ways that are unanticipated and possibly harmful to achieve the tasks. The new code of conduct tackles the issue directly, stating Microsoft's AI will not take “excessive risks” to finish any particular project.
Public Feedback and the Road to Training Future Models
Microsoft is not considering this code of conduct as complete. The company is asking for public comment on the document over the next six week period before it is finalized. Some of the most pressing questions are: How much should AI respect user boundaries and how should it interact with emotionally sensitive users?
After the feedback period ends, Suleiman stated that the code will be directly used in the training of the models that Microsoft will build. This means that the principles in the document will not be just stated as guidelines, but should be internalized by how Microsoft's AI systems think or respond.
A Call for Industry-Wide Coordination on AI Safety
Microsoft's announcement of its AI code of conduct follows similar public appeals by Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman for the AI industry to take its development "with a sense of responsibility" for the sake of safety. Suleiman agreed, stressing the need for stronger cooperation between AI labs in order to maintain control over the increasingly powerful technology.
“It's a good time for everyone to have this discussion and breathe,” Suleiman said.
AI systems are increasingly capable and autonomous, and with this, structured governance can become a standard practice in the industry, such as Microsoft's code of conduct, which will help to ensure that human oversight continues to play a key role in the development of AI.
