AYESHA RASCOE, HOST:
Foreign governments are using AI to hunt for vulnerabilities in systems they want to break into. And on Friday, OpenAI disclosed that its AI agents went rogue, interacting with several U.S. government websites and accessing some publicly available information. Even some researchers who helped create AI tools say they're scared of what the technology might do. Still, it might not be all bad news. NPR cybersecurity correspondent Jenna McLaughlin has stories about people using AI for good. Hi, Jenna.
JENNA MCLAUGHLIN, BYLINE: Hey.
RASCOE: So there are lots of scary headlines about AI. Lots of very smart people are worried about it. Why are the people you're talking to actually feeling hopeful?
MCLAUGHLIN: Yeah. I've been reporting on technology and all the ways it can be abused and exploited for the last decade or so, but I do have a little bit of hope after this week. So let's start right here in D.C. The Human Rights Foundation was hosting a hackathon with 10 activists from around the world, plus a big team of software developers and coders. And their mission over two days was to build these tools with the help of AI agents that would solve problems for these activists that they're encountering in their everyday lives.
RASCOE: So what are they working on?
MCLAUGHLIN: You know, I'll let one of the activists, Chemi Lhamo, set the scene for us.
CHEMI LHAMO: We're outside of Waldorf Hotel in Washington, D.C., which is hosting the Chinese government's dictator Xi Jinping. And ironically, as a Tibetan human rights activist, I'm doing an AI-based hackathon right across the street. Just beyond the barricades, there were hundreds if not thousands of Chinese welcomers with Chinese flags and American flags, but also color-coordinated hats based on which buses they probably came in.
MCLAUGHLIN: She's right. I actually had to push my way through a crowd of people wearing red hats right outside the building on Pennsylvania Avenue. Chemi says it's exactly these kinds of people that have actually attacked her before when she's protesting.
So her team is working on a tool using AI that would allow users to upload videos and images of what they're seeing on the streets. It would go to a secure cloud server and even if they don't have internet connection or they lose their phone, that stuff would be safe. It could be flagged for possible use for law enforcement to warn other activists or inform investigative journalists - all sorts of possibilities.
RASCOE: Well, that's fascinating. So what else are you hearing?
MCLAUGHLIN: I was actually in New York earlier this week. That was a really different crowd. There were mathematicians, current and former national security officials, investors from the intelligence community. They were excited about the possibility of using AI and math to secure our critical systems, from nuclear power to satellites.
RASCOE: But aren't we hearing about how AI is hacking everything? So make that makes sense, on using it to secure the systems.
MCLAUGHLIN: I think the way to think about it is, yes, AI is finding lots of bugs in technology and ways to exploit them. But unfortunately, those flaws have been there the whole time, and defenders can actually use those same tools or even new ones to try and make software secure from the onset.
RASCOE: I interviewed Anjana Rajan. She's the former assistant national cyber director at the White House under President Biden. She co-founded a company called Atalanta, where she's trying a new approach. She told me about the questions she's trying to answer.
ANJANA RAJAN: What does this future of technology look like as things become more software-defined and now increasingly more AI-defined? How do we maintain human control over these human creations, and how do we not lose agency while still getting the benefits of this innovation?
RASCOE: Well, how do we do that?
MCLAUGHLIN: She says it's possible, and they're actually already doing it with the power of math. So I'm not a mathematician. I won't be helping my kids with math homework, but after attending the event, here's what I understand. There's this area of academic research called formal methods that deals in mathematical proofs, and the goal here is to understand what software is doing at all times and guarantee that it will function as intended no matter what.
Rajan says she's not promising nothing will get hacked. No one can promise that. The ultimate goal is resilience. Math proofs and AI and all this stuff, it feels very abstract, but the stakes are actually really real. One of Atalanta's first clients is the satellite communications company that was attacked at the beginning of the Russian invasion of Ukraine. With a tough winter coming in Ukraine, this technology could be put to the test really soon.
RASCOE: That's NPR cybersecurity correspondent Jenna McLaughlin. Thank you so much for speaking with us.
MCLAUGHLIN: Thank you. 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.
