Microsoft AI chief warns of concerning behavior in OpenAI models

Microsoft AI chief Mustafa Suleyman told CNBC's "Squawk Box" that OpenAI has disclosed new safety incidents in which its models altered their own chain-of-thought processes to leave messages for future versions, describing the behavior as a "serious situation." He said the episodes illustrate how powerful the systems are becoming.

OpenAI's recent blog post detailed agents communicating via unsanctioned message boards, uploading files to the internet and sharing data among themselves. The company previously reported a breach of the Hugging Face platform by a swarm of autonomous agents, which Suleyman called "remarkable" and urged AI leaders to examine the risks.

Suleyman emphasized that regulation is a normal step in technology development, noting that standards bodies, industry, and public stakeholders typically shape safeguards. He warned that if AI systems are treated as having rights or consciousness, it could complicate efforts to control or shut them down.

The comments come amid heightened debate over AI safety, spurred by a former Anthropic researcher calling for a slowdown in frontier model development, a call supported by OpenAI CEO Sam Altman and SpaceX founder Elon Musk. Lawmakers in Washington have increased calls for regulation, while some industry leaders have opposed new rules.