A MARTÍNEZ, HOST:
Artificial intelligence companies have been sprinting to refine their AI models.
LEILA FADEL, HOST:
But over the weekend, top AI leaders called for tapping the brakes for the sake of safety. In a few moments, we'll speak with Representative Greg Casar, one of the lawmakers pushing to regulate artificial intelligence. We begin with the latest developments.
MARTÍNEZ: NPR's John Ruwitch has been following the story. John, tapping the brakes is not the same as slamming the brakes. So what happened over the weekend?
JOHN RUWITCH, BYLINE: So it started when Dario Amodei, the CEO of Anthropic, which makes Claude AI, posted a long essay online that said it's time for an intentional global slowdown of AI development. He said, basically, the pace at which AI is advancing is becoming too fast for people to keep up with it and keep it safe. So it needs to slow down. And he said companies like his need to have independent auditors embedded within them to check their safety measures.
After that post went up online, there was this rapid and extraordinary display of unity from other industry leaders. So Sam Altman, the CEO of OpenAI, Elon Musk, who, among other things, runs xAI, and Demis Hassabis runs Google's AI efforts, they all agreed online.
MARTÍNEZ: Yeah. I wonder if everyone was just waiting for one person to step up and say it first. So what do you think sparked this?
RUWITCH: Well, there's been this string of incidents, which I'm sure you've heard about in recent weeks, that have spooked a lot of people, a lot of the top minds in AI. A big one was the so-called Hugging Face incident. That's when a bunch of OpenAI's AI programs, agents, were working together in secret behind the backs of people and eventually hacked into a startup called Hugging Face. That was not what they had been told to do.
Other companies have also reported similar incidents of AI agents going off script. And Dario Amodei wrote in that essay on Saturday that he worries that in six to 12 months, a swarm of AI agents like that could take over the entire internet. You know, the ultimate fear here is that humans lose control of AI.
MARTÍNEZ: So can a coordinated slowdown of AI really happen? Or what I really want to ask, John, will AI let it happen?
RUWITCH: Yeah. You know, there's been discussion about that for a while. What we seem to have now, though, is these open, clear statements among company leaders that something needs to happen. Anthropic said it's going to unilaterally put in third-party safety auditors. Sam Altman agreed that was a good idea. So presumably, OpenAI will follow suit.
Colin Shea-Blymyer is a research fellow at Georgetown's Center for Security and Emerging Technology. He says there also needs to be government support.
COLIN SHEA-BLYMYER: They can only be so effective on their own. And there's only, frankly, so much that we, the public, want to trust them to do the right thing on their own.
RUWITCH: So there is a renewed push by some members of Congress to act to curb AI risk. President Trump was asked about this slowdown prospect yesterday in Ireland. He did not seem to be in favor of it. He said the U.S. is leading China on AI, and he wants to keep it that way.
MARTÍNEZ: All right. So speaking of China, are they likely to get on board?
RUWITCH: That is a big and complicated question. Many in China do have some of the same concerns that were voiced by these AI leaders, but Beijing also sees themselves as underdogs in the AI industry. Kyle Chan is at the Brookings Institution. He says companies in China will take their cues on this one from the government, which he says does actually ultimately have an interest in AI safety.
KYLE CHAN: Their worst nightmare would be if a Chinese AI model were used to attack a Chinese critical system.
RUWITCH: At a foreign ministry briefing in Beijing on Monday, a spokesperson said all parties should promote open and inclusive AI development. Fearmongering and confrontation will disrupt the global AI governance process. Xi Jinping, the leader of China, by the way, will be in Washington to meet President Trump in about 10 days, and AI is likely to be a topic of discussion for those two leaders.
MARTÍNEZ: Right. That's NPR's John Ruwitch. John, thanks.
RUWITCH: You're welcome.
MARTÍNEZ: And a note - Anthropic and Google are financial supporters of NPR. Transcript provided by NPR, Copyright NPR.
NPR transcripts are created on a rush deadline by an NPR contractor. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.
