SCOTT DETROW, HOST:
The AI company Anthropic claims a new model it introduced last month is so powerful, it is too dangerous to release to the public. It's called Claude Mythos, and it has raised concerns from top banking officials. Our colleagues Darian Woods and Wailin Wong over at The Indicator ask, are bank accounts still safe in this era of rapidly advanced AI?
DARIAN WOODS, BYLINE: The new AI model, Claude Mythos, is, in some ways, a typical language model. It wasn't fundamentally different from a chatbot trained to answer general questions or write lines of code, like Anthropic's other Claude models or ChatGPT.
WAILIN WONG, BYLINE: But when Anthropic was testing Claude Mythos, people in the company, like Michael Moore, started getting concerned. He heads up the company's cybersecurity products.
MICHAEL MOORE: It showed a really advanced capability of chaining together individual vulnerabilities into an exploit that was potentially really dangerous.
WONG: He means that Mythos could find multiple lines of code that on their own weren't hugely problematic but together could leave the software exposed. It's like if a thief could break into your apartment block's front gate and your building's front door and your apartment unit's lock.
MOORE: It was the first time that, if we just were to release the model out into the public, we were concerned that the scale of AI would allow offensive actors to actually go break into systems at an unprecedented rate.
WOODS: So Anthropic gave special access to Mythos to only a small group of companies. This was to allow them to fix their software before Mythos had a wider release. Anthropic called it Project Glasswing.
WONG: At first, there was only one named financial institution given early access, JPMorgan. Later, Bank of America was, too. That's two out of America's 4,000 banks.
WOODS: Should people be worried about their life savings potentially disappearing from their bank accounts?
MOORE: I'm really excited to see the work that's being done to patch as many vulnerabilities as possible so that it doesn't come to a moment of strong concern for most people. Hopefully, it's a little bit kind of like what Y2K turned out to be, which was really not much.
WOODS: This drive to help patch up the world's software is partly why Anthropic released its next product. Later in April, Claude released a special tool just for finding and fixing software vulnerabilities.
WONG: A skeptic might say this is convenient timing.
WOODS: What do you say to critics who might argue that Anthropic is creating the problem it's charging people money to solve?
MOORE: It's a good question. Cloud security is really about looking at code bases that have been around for decades. Those kind of bugs are the things that we expect AI to find in the future, and that's why patching them using AI is so important now. As we move and continue to evolve as an industry and more code is written by AI, we definitely expect the security of the industry to get better.
WONG: But other cybersecurity experts aren't as confident. Rachel Greenstadt is a computer science professor at New York University. She points out that less experienced developers are vibe coding software.
RACHEL GREENSTADT: People are able to write much more complex things much quicker, that they don't understand, that the AI doesn't understand, and this is an opportunity for whole new classes of bugs.
WONG: There are AI models available out there that are already roughly as capable as Claude Mythos. In fact, an open model found of bug lurking in an old operating system.
WOODS: To Rachel, we are in a new, faster era but not one that brings challenges that are fundamentally different from what we've tackled before.
WONG: Wailin Wong.
WOODS: Darian Woods, NPR News.
(SOUNDBITE OF J DILLA'S "REQUIEM") 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.
