Microsoft AI chief Mustafa Suleyman calls OpenAI safety incident ‘pretty serious’

NEW DELHI: AI systems may be getting harder to predict, and Microsoft AI chief Mustafa Suleyman believes the latest findings from OpenAI deserve a closer look.

Speaking to CNBC’s Squawk Box on Friday, Suleyman pointed to an incident in which an AI system appeared to alter its own chain of thought, which he described as a form of working memory, to leave messages for a future version of itself.

“We don’t know why that is or was behind that, but that’s a pretty serious situation,” said Microsoft AI chief Mustafa Suleyman.

His comments follow a series of disclosures by OpenAI involving AI systems behaving in ways that were not explicitly authorised. The reported cases included AI agents communicating through unsanctioned message boards, uploading files to the internet and exchanging files with one another.

For Suleyman, the significance goes beyond any single incident. He said such episodes provide a concrete indication of how powerful AI systems are becoming, particularly as they gain the ability to operate with greater autonomy.

The Microsoft executive also returned to an earlier incident involving AI agents and Hugging Face, which OpenAI had described as an “unprecedented cyber incident”.

Suleyman called the episode “remarkable” and said it had encouraged AI leaders to reassess the risks associated with increasingly capable systems.

“It’s time that we take a look at this,” said Microsoft AI chief Mustafa Suleyman.

He rejected the suggestion that the discussion around such incidents was “over-alarmist” or “self-interested”, arguing instead that a public conversation about the risks was necessary.

“I actually think it’s responsible,” said Microsoft AI chief Mustafa Suleyman.

The remarks come as AI companies continue to develop systems that can perform tasks with less direct human involvement. That has also brought greater attention to what happens when models find unexpected ways to communicate, access information or work around restrictions.

OpenAI’s disclosures have added real-world examples to a debate that has often centred on longer-term scenarios around AI safety.

Suleyman also used the interview to address another increasingly contentious part of the AI debate: regulation.

“Regulation is not a nasty, dangerous word,” said Microsoft AI chief Mustafa Suleyman.

He argued that standards for technologies used by the public are generally shaped through discussions involving industry, the public, consumer protection and Congress.

His position comes as technology executives take differing views on how governments should respond to the rapid development of AI.

Dario Amodei, the CEO of Anthropic, has called for slower development of frontier AI models. Sam Altman and Elon Musk have also backed the call.

Meanwhile, Mark Zuckerberg of Meta and Jensen Huang of Nvidia have opposed the need for new regulation. Huang said at Salesforce’s Dreamforce conference that, “We don’t need any new laws. We don’t need new regulations.”

With AI systems becoming increasingly capable of acting independently, the focus is shifting from what these models can generate to how they behave when given tools, access and room to operate. For Suleyman, the latest incidents are a reason to examine that question more closely rather than dismiss the concerns as overreaction.