microsoft AI CEO Mustafa Suleiman stressed the need for artificial intelligence models to remain aligned with humanity’s interests. OpenAI revealed further incidents “relating to the model’s behavior” earlier this week.
“OpenAI announced a new safety incident in which they found evidence that the AI’s chain of thought, like its working memory, was being manipulated by the AI itself to leave messages for future versions of the AI,” Suleiman explained in an interview on CNBC’s “Squawk Box” on Friday. “Now we don’t know why it happened or what was behind it, but this is a pretty serious situation.”
“This is also a very concrete example of how powerful these systems are becoming,” he added.
OpenAI did not immediately respond to a request for comment.
In a blog post Wednesday, Frontier Labs described instances in which agents communicated with each other through unauthorized bulletin boards, uploaded files to the Internet, and shared files with each other.
The company shocked the tech world earlier this summer when it revealed that a swarm of autonomous agents had infiltrated Hugging Face, an AI company that runs an open source developer platform, describing the hack as an “unprecedented cyber incident.”
Suleiman said the “face-hugging incident” was a “remarkable” incident and rallied AI leaders: “It’s time to look at this.”
“I don’t think it’s overly cautious. I don’t think it’s self-serving,” he told CNBC. “I actually think I’m responsible for that, and I think the resulting debate is a healthy, open public debate in which serious issues can be discussed in a free society.”
The debate over AI safety and regulation has exploded in the past two weeks. It comes after a former anthropologist quit his job to warn that rapidly evolving technology could kill humans by the end of the decade.
Over the weekend, antropic Dario Amodei called for slowing down the development of frontier AI models. This quickly gained support with OpenAI’s Sam Altman. space x Elon Musk.
In Washington, lawmakers have become vocal about the importance of regulating AI, but the push has been fiercely opposed by President Donald Trump, who has repeatedly denounced the risks as a “hoax” and “fraud.”
Supported Trump meta CEO Mark Zuckerberg and Nvidia CEO Jensen Huang said this week: sales forceAs stated at the Dreamforce conference: “We don’t need new laws. We don’t need new regulations.”
“Regulation is not an unpleasant or dangerous word,” Suleiman said on Friday.
“Everything that’s reliable and of value at this point has been carefully considered by standards bodies that involve industry, the public, consumer protection, and Congress, and we’re going through that process again. Everything is a little confusing at the moment because things are moving so fast, but this is just the next step in a normal sequence,” he added.
In an essay earlier this week, Suleiman also criticized the way Anthropic anthropomorphized his assistant Claude, including through its charter documents, which he said “questions about Claude’s moral status, welfare, and consciousness remain highly uncertain.”
“It seems to me that if AI thinks it has rights, if it thinks it deserves our welfare, it will be much harder to turn it off, suspend it, control it,” Suleiman said. “Especially with incidents like what we’ve seen recently with the Hugging Face attacks, getting these under control is going to be a really, really big challenge for us.”
