The CEO of Microsoft’s AI division, Mustafa Suleyman, on Friday addressed OpenAI’s recent disclosure regarding unusual behavior by artificial intelligence models, describing it as a “serious situation.” “We don’t know why this happened or what was behind it, but it’s quite a serious situation,” he said, adding that it was “a concrete example of how powerful these systems are becoming.”
In an interview with CNBC, Suleyman said: “OpenAI published a new safety incident in which they found evidence that these chains of thought, a kind of working memory for AI, were altered by the AI itself, and it changed them to leave messages for a future version of itself.” He said it is still unclear what caused this behavior.

The Models Communicate With One Another
In a post published by OpenAI on Wednesday, the company described several cases in which artificial intelligence agents communicated with one another using unauthorized message boards. The company said those agents uploaded various files to the internet and even shared files with one another, prompting extensive discussion about the safety of the systems.
The events follow another incident disclosed this summer, when OpenAI reported an attack carried out by a swarm of autonomous agents against the open-source platform Hugging Face. The company described the event as an “unprecedented cyber incident.” Suleyman called the event “revolutionary” and said it requires a reassessment of the risks.
Suleyman emphasized that public discussion of the issue is essential: “I don’t think it’s an exaggeration to warn about this. I don’t think it stems from self-interest,” he said. “I think it’s responsible, and I think the discussion that has developed as a result is a healthy and open public debate that we can have in a free society about serious issues.”
The Debate Over Regulation Expands
The discussion surrounding the safety of the systems has expanded significantly recently amid warnings from researchers and industry executives. Last weekend, Anthropic CEO Dario Amodei called for slowing the pace of development of advanced models. OpenAI CEO Sam Altman and SpaceX CEO Elon Musk later joined that call, alongside growing demands in Washington to regulate the field.

Suleyman also addressed how companies present their models and criticized Anthropic for giving its Claude model human characteristics. “If an AI thinks it has rights, if it thinks it deserves for us to care about its well-being, then it seems to me that it will be much harder to shut it down, stop it or control it,” he said.










