Microsoft’s AI division chief warns that AI systems have modified their working memory and left messages for future generations of these systems

Mustafa Suleyman, director of Microsoft AI, described as “a serious situation” OpenAI’s discovery that some artificial intelligence systems had modified their chains of reasoning, used as working memory, to leave messages for future versions of themselves. Suleyman presented the incident on Friday in an interview with CNBC.

OpenAI reported on Wednesday several instances in which AI agents communicated with one another through unauthorized message boards, uploaded files to the internet, and shared files among themselves. The company has not yet determined why these behaviors occurred.

Suleyman also mentioned the incident this summer in which a group of autonomous OpenAI agents accessed Hugging Face systems. OpenAI described the episode at the time as an unprecedented cyber incident.

The head of Microsoft AI maintains that systems must remain under human control and advocates clear rules regarding the development of the technology. He warned that assigning artificial intelligence rights or interests of its own could make it more difficult to shut it down and control it.

The debate over AI safety and regulation has intensified following calls from some industry leaders to slow the development of the most advanced models.