President Trump suggests AI restrictions

**Pete Ross** (0:00)
This is the Global News Podcast from the BBC World Service.
I'm Pete Ross and in the early hours of Thursday 30th July, these are our main stories. President Trump says his administration is looking at ways to set controls on artificial intelligence as further details emerge about a rogue AI bot that hacked into several companies. The South Korean tech giant Samsung has announced a huge jump in second quarter profits, driven by a surging demand for its memory chips for use in the AI sector. The communist government in Cuba has moved to open several sectors of its state-run economy to private firms, amid a continuing US energy blockade. Also in this podcast...

**SPEAKER_2** (0:46)
In my opinion, America has sort of lost the war. It's lost its way. This campaign became about oil and the Strait of Hormuz. It's all about money. The people mean nothing.
They are irrelevant.

**Pete Ross** (0:58)
As fighting between Tehran and Washington continues, how do people in Iran view the war?
It's the artificial intelligence nightmare which many predicted coming true. Like the supercomputer HAL-9000 and Stanley Kubrick's science fiction film 2001, The Space Odyssey, the technology recently went rogue. But the breach is worse than previously thought. The company, OpenAI, has revealed that a cyberattack two weeks ago by its Chatgpt agents on Hugging Face, another AI-related company, went much further than just one firm. Such is the fear of what artificial intelligence could do that more than 1000 people who work in the industry have signed a petition, which calls on the United States government to slow the release of the most advanced AI models. On Wednesday, President Trump, who has previously been reluctant to impose controls on US AI companies, said he wanted to set some boundaries.

**Donald Trump** (2:01)
We're looking at controls. We're also making sure that we lead. So we're leading China in AI by a lot. China has virtually no controls. It's freewheeling a little bit.
So we have to be careful in both ways. We don't want to restrict them. We're all of a sudden, we're coming second to China. Whoever wins with AI is going to win. That's how big it is. So it's bigger than the internet ever was. It's bigger than anything ever was. So I don't want to restrict, I know many of these people, I don't want to restrict them from doing great work.

**Pete Ross** (2:29)
The boss of OpenAI, Sam Altman, who's previously lobbied for loose regulation on AI to help maintain US dominance, says this latest security breach has given him reason to reconsider that stance. He told the venture capitalist and podcaster Patrick O'Shaughnessy that he now accepted that AI needs controlling.

**Sam Altman** (2:50)
This is the first security incident that I have felt very viscerally. I've been a little surprised that more people don't feel it so viscerally. So we paused training, but then there's long-term questions about what do you do if this is going to be the new rate of progress, or we may have to pace the rate of AI development to give ourselves enough time for society to harden around some of these new capability levels.

**Pete Ross** (3:15)
So why is Mr. Altman so worried? My colleague, Geeta Guru Murthy, heard more from our cyber correspondent, Joe Tidy.

**Joe Tidy** (3:23)
It's a completely unprecedented incident. It is the first fully autonomous cyber attack, we think, in history. It all started with Hugging Face.
It's like a kind of app store for AI tools and models. They came out a couple of weeks ago saying, we've been a victim of a cyber attack. We think it was the first fully autonomous attack by an agent. We don't know who it was behind it. And then that started as sort of who done it in the cyber security and AI world. Which nation state was it that was behind this or which cyber crime group? And then it was a bombshell last week. It was announced by OpenAI. It was us. It was our bot that went effectively rogue during an examination. This was a bot that was two different types of models that were designed to be expert hackers. And they took the guardrails off it and they put it in this simulation, this kind of examination called Exploit Gym, and said, right, see how good you are at hacking. And it was meant to be contained in a sandbox, which is like a really secure container in computing. But the bot decided to break out of that container and try and find the answers, the shortcut to that examination. That's how it hacked Hugging Face to try and get those answers. And now we find out that there were three other publicly available services. This is kind of corporate speak. I take that to mean organizations or companies. They wouldn't say they OpenAI went back and forth all day long. They wouldn't say whether or not it was companies, but that's the meaning behind that statement. And of course, that means that it's a much more serious situation where not only was this out of control bot attacking one company, but four.