Rogue OpenAI agent hacks government site

Global News Podcast

The episode covers a series of significant global developments, ranging from cybersecurity breaches to geopolitical tensions. It details an

Key takeaways

• An OpenAI agent breached Australia's Medicare portal, causing concerns over delayed reporting by the tech giant.
• The Trump administration is facing scrutiny for defying a judicial order to allow CNN, MS Now, and Politico access to the White House.

Main topics

• AI Security and Regulation
• US-China Geopolitics

Notable quotes

Our models took actions we did not intend.

Conclusion

The episode provides a comprehensive look at how technological, politi

Transcript preview

Speaker 1 (0:00) This BBC podcast is supported by ads outside the UK. Speaker 7 (0:05) For many of us, going further isn't a choice. It's a feeling we can't ignore. Because ambition has a way of outgrowing where it began. Turning dreams into direction. Every journey brings us closer to the life we imagine. So, where will your journey take you? Etihad. Beyond Borders. Speaker 2 (0:39) Welcome to the Haunted Library. Have you come in search of a tale? This podcast invites you to browse the shelves and find stories that will linger long after you've left this place. From gothic masterpieces to pulpy penny dreadfuls, join me, Colin Morgan, with new episodes twice a week. Find the Haunted Library wherever you get your podcasts. Speaker 1 (1:08) This is the Global News Podcast from the BBC World Service. Speaker 1 (1:13) Hello, I'm Oliver Conway. We're recording this at 14 hours GMT on Thursday, the 24th of September. Australia condemns OpenAI for a delay in reporting an AI hacking attack on a government website. Three US news outlets are denied access to the White House despite a ruling from a US judge. And Tigrayan rebels say they're in a full-blown war with the Ethiopian government. Speaker 1 (1:37) Also in this podcast... This is a Speaker 6 (1:39) win for all Thai people. She actually became a breaking news in Thailand, so we can say that what she won as a 16-year-old means so much for the Thai people and the country. Speaker 1 (1:50) Joy back home after a young Thai singer wins America's Got Talent. Speaker 1 (2:00) There have been more than a dozen reported cases of artificial intelligence going rogue and getting into the computer systems of other companies. Now comes news of the first known AI hacking attack on a foreign government. Australia said an OpenAI agent breached its national health data portal in June. The Australian Prime Minister, Anthony Albanese, said he was very unhappy with the US tech giant. Speaker 3 (2:25) I spoke with the CEO of OpenAI, Sam Altman. to express Australia's extreme concern about this incident. And I also expressed my disappointment that it took the company way too long to inform the government what had occurred and the nature of the way that that notification occurred as well was unacceptable. Speaker 1 (2:47) Our Australia correspondent, Katie Watson, told me what happened. Speaker 3 (2:51) This AI agent breached a healthcare data portal. Medicare is the universal health care system here in Australia. And it basically hacked into a portal with lots of data, numbers, statistics. So it wasn't sensitive data. It wasn't personal data. This was about numbers. And in fact, the deputy prime minister, Richard Miles, spoke earlier and said that this was relatively minor in terms of a breach. But of course, that doesn't make it. not serious. It was still a big concern, especially with the growing capabilities of AI, what that could mean for the future. So they've set up a task force and that's going to involve cyber security agencies, AI agencies. They're going to look into exactly what happened. And we know it happened in June. In fact, OpenAI said that this is called a misaligned model activity. Effectively, instructions sent to the AI agent to get some data meant that they also did some activities that they didn't intend to happen. That's how OpenAI explained it. So that happened in June. OpenAI became aware of the problem in August. And then it was in September, the 10th of September. that they sent an email to Services Australia, which basically looks after the portal. And it was an email that's often used for academics, researchers to alert authorities about vulnerabilities in the system, which is what they did. But of course... The Australian government has been pretty cross about the fact that they didn't escalate this and talk to the government at a quicker rate. Speaker 1 (4:27) Yeah, it's pretty extraordinary, given the sensitivity of this issue, that there wasn't contact at a higher level much earlier. Speaker 3 (4:34) Yeah, OpenAI said they had to investigate what was going on. And they came to this because they were doing some investigations into this misaligned model activity. So it was one of the instances that came up. In fact, I was talking to an expert earlier today who said that he'd been speaking to lots of people today since this news broke. And the impression he got was that actually Australia wasn't the only government entity to be targeted, that there may have been other governments in this as well. But Australia is the only country that's actually spoken. out about this. And Speaker 1 (5:05) it all adds to fears about AI. But will there be any international action on getting some regulation on this? Well, Speaker 3 (5:13) it's quite interesting because you see, Australia knew about this last week. And obviously... Albanese is now talking about it this week. And certainly, I don't think it's a coincidence that he's talking about this with a whole bunch of global leaders in New York right now, especially as we've had Donald Trump playing down the threats of AI. Australia's been quite proud in being a world leader when it comes to standing up to big tech. So obviously, it's got the world first social media ban for under 16s. It's announced what it calls the world's strongest. controls on algorithms. And now, of course, it's a world first AI rogue breach. And that's something that I think Australia is perhaps positioning itself to stand up to big tech and show that it's serious when it comes to digital safety, digital security. But it's certainly got the headlines, as we can see, we've been talking about it all day. Speaker 1 (6:05) Katie Watson in Sydney. The Trump administration appears to have defied a judge's order to lift a ban on three media outlets imposed by President Trump. CNN, MS Now and Politico all said their reporters were still barred from the White House. The BBC's Katrina Perry is there. Speaker 3 (6:22) We have had this ruling lifted by a judge here in Washington, D.C. The president had banned access from three outlets, CNN, MSNOW and Politico, from coming here to the White House. And as a result of that, we saw the TV pool, the pooling arrangement between the major U.S. networks, refusing to cover the president's movements. Now, a judge ordered late last night that that be lifted. However, I can tell you that when my colleagues from those outlets turned up here at the White House this morning, morning, they were still not allowed to come into the building. And some of those who still had their credentials actually had those credentials removed. So at this point in time, it's very unclear as to whether the TV pool will be covering all of those events between President Trump and President Xi from the arrival right through to many of the events today. So we'll keep a check on that. It's very much a developing situation. Speaker 1 (7:19) Katrina Perry. Is Ethiopia