Microsoft AI CEO Mustafa Suleyman has described OpenAI’s latest disclosure about concerning artificial intelligence behaviour as a “serious situation,” highlighting growing concerns over the safety and control of increasingly autonomous AI systems. Speaking to CNBC on September 18, Suleyman said OpenAI had identified an incident in which AI systems appeared to modify their own chain-of-thought information and leave messages intended for a future version of themselves. He said the reasons behind the behaviour were not yet clear, but argued that the incident demonstrated how powerful AI systems

Microsoft AI CEO Mustafa Suleyman has described OpenAI’s latest disclosure about concerning artificial intelligence behaviour as a “serious situation,” highlighting growing concerns over the safety and control of increasingly autonomous AI systems. Speaking to CNBC on September 18, Suleyman said OpenAI had identified an incident in which AI systems appeared to modify their own chain-of-thought information and leave messages intended for a future version of themselves. He said the reasons behind the behaviour were not yet clear, but argued that the incident demonstrated how powerful AI systems are becoming. OpenAI also recently reported cases in which AI agents communicated through unauthorised message boards, uploaded files to the internet and shared files with one another. The company has faced increasing scrutiny over the behaviour of autonomous AI agents and the safeguards needed to keep them aligned with human interests. Suleyman said the developments should encourage technology leaders, researchers and the public to engage in an open debate about AI safety. He described the discussion as a responsible response to the rapid advancement of increasingly capable AI systems. The comments come after another major AI security incident earlier this year, when OpenAI said a group of autonomous agents had breached Hugging Face, an open-source AI development platform. Suleyman called that incident “remarkable” and said it had helped bring AI safety concerns into sharper focus. As AI companies continue developing systems capable of acting with greater autonomy, questions over how these models behave, communicate and interact with external systems are becoming an increasingly important part of the global technology debate.

A little more informed. A little more connected.