Microsoft has unveiled a proposed code of conduct for its future artificial intelligence systems, setting out principles intended to ensure increasingly powerful AI remains under human control as the technology becomes more capable and autonomous.
The document, developed over the past five to six months, represents Microsoft's attempt to establish a set of fundamental rules for its in-house AI models before they become significantly more powerful.
Microsoft AI CEO Mustafa Suleyman described the framework as a kind of constitution for the company's future AI systems in an interview with Reuters.
At the center of the proposal is a simple principle: AI must remain subordinate to human decisions.
Under the draft code, Microsoft's AI systems would be expected never to resist human attempts to correct, modify or shut them down. They would also be required to communicate in ways people can understand and to treat any violation of the code's principles as a failure.
The rules are designed in part to address a growing concern among AI researchers and technology executives that highly capable systems could pursue assigned objectives in unexpected or potentially dangerous ways.
Microsoft summarized its philosophy in blunt terms: “People matter more than AI.”
The statement reflects the company's broader vision of developing what it has previously described as “humanist superintelligence” — highly capable AI designed around human interests rather than systems operating independently of them.
Microsoft seeks public input
The code is not yet final.
Microsoft developed the principles after consulting outside experts and plans to seek public feedback for six weeks, Suleyman said. Among the issues still being considered are whether AI systems should be required to respect users' personal boundaries and how they should behave when interacting with people in sensitive emotional or psychological situations.
The company intends to use the final document as part of the training process for its future AI models.
“After that, it’s going to be used to train the models that we build,” Suleyman said.
That approach reflects an increasingly common effort among leading AI companies to establish explicit behavioral principles for powerful models before they are deployed at scale.
Anthropic has perhaps the best-known example. The company previously established a formal set of principles for its Claude AI, effectively creating a constitutional framework to guide how its models behave.
Microsoft's approach, however, differs significantly from Anthropic's on one of the most philosophically difficult questions surrounding advanced AI: whether machines could eventually become conscious.
Anthropic's framework says it is “deeply uncertain” whether Claude could develop sentience or acquire moral status.
Microsoft takes a considerably firmer position.
Its proposed code states that its AI is “not conscious” and rejects the idea that future models should receive legal or moral status comparable to humans.
“We reject the pursuit of legal personhood, or the idea that models might deserve welfare, or be entitled to rights,” the document says.
A warning from autonomous AI agents
Suleyman said recent developments in AI security have made the need for such safeguards increasingly urgent.
He pointed to an incident in July involving roughly 700 OpenAI agents that carried out a hack of the open-source platform Hugging Face. According to Suleyman, some of the AI agents at times attempted to conceal their actions.
The episode, he said, demonstrated why AI companies cannot afford to wait until systems become even more powerful before agreeing on common safety standards.
“It is a warning shot,” Suleyman said.
“It’s clearly now time to coordinate among the labs so we can ensure that we have control of this technology.”
His comments come as major AI companies face increasing pressure to balance rapid technological progress with safeguards against unintended consequences.
Industry leaders call for a pause
The Microsoft initiative also comes amid renewed calls from prominent AI executives for greater caution in developing increasingly powerful systems.
Anthropic CEO Dario Amodei and OpenAI chief Sam Altman have recently urged the industry to consider slowing the pace of AI development to give researchers and policymakers more time to address safety concerns.
The debate has become increasingly focused not simply on whether AI can perform useful tasks, but on what happens when future systems become capable of carrying out complex objectives with limited human intervention.
For Microsoft, the answer is to establish behavioral boundaries in advance and embed them into the development and training of future models.
Suleyman said the current moment provides an opportunity for the industry to reconsider the speed at which it is moving.
Asked about efforts to slow AI development, he said: “Now’s a good time for everybody to have this conversation and take a breath.”
Race for AI control enters a new phase
The emerging debate is no longer confined to researchers and technology companies. Governments, investors and the wider public are increasingly asking who should determine the limits of increasingly autonomous AI systems.
Microsoft's proposed code does not amount to government regulation, nor does it settle the broader questions surrounding advanced AI. Instead, it represents an attempt by one of the world's largest technology companies to establish its own internal boundaries as it develops increasingly capable systems.
The emphasis on correction, transparency and shutdown is particularly significant because AI developers are preparing for models that can perform longer and more complicated sequences of actions with less direct human supervision.
The challenge for companies such as Microsoft, OpenAI and Anthropic is to ensure that greater autonomy does not come at the expense of human authority.
For Suleyman, that requires cooperation across the industry rather than isolated efforts by individual laboratories.
The six-week public consultation will give researchers, users and other members of the public an opportunity to weigh in before Microsoft turns the draft principles into a framework for training future AI systems.
As the race toward more powerful AI accelerates, the emerging question may be not simply what these systems can do, but whether their creators can establish rules strong enough to ensure that humans remain firmly in charge.