Cyberattacks reach a new milestone

Editor's note: The following text is a transcript of a podcast story. To listen to the story, click on the arrow beneath the headline above.

NICK EICHER, HOST: Coming up next on The World and Everything in It: the next frontier of the cyberattack.

Last month, AI developer Anthropic said it detected the first known cyber-espionage campaign … orchestrated by A-I. Here’s WORLD Radio Executive Producer Paul Butler.

PAUL BUTLER: Movies tend to portray computer hacking like this: a shadowy figure in a black hoodie taps a few code words into a computer. Within moments, the criminal accesses the secret files.

VICTOR BENJAMIN: But in reality, hacking is a very slow and just a trial and error process. So you kind of like knock on multiple doors, see what’s open, what you can crack open to get further into a system.

Victor Benjamin is an assistant professor at Arizona State University’s Department of Information Systems. He says that trial and error work has gotten easier, as artificial intelligence allows hackers to automate the manual work of hacking and scale up their attacks.

Using large language models—or LLMs—hackers can quickly research targets and spot an organization’s security gaps.

BENJAMIN: If you find something that may be a vulnerability, like it's running a outdated server side software or something, then you can use an LLM further to write code to target that vulnerability. So the LLM and how the attackers used it, it basically is a kind of like a partner in the whole cyber attack chain.

Software company Anthropic believes that China used its Claude tool to attack about 30 American targets. Hackers overrode the software’s security features, a process known as jailbreaking. They told the chatbot they belonged to a cybersecurity firm and used Claude to research the intended targets. Claude organized the information and created malware to take advantage of the weak points in each target’s systems. While hackers only successfully compromised a handful of victims, Benjamin says the attacks mark a troubling technological milestone.

BENJAMIN: So this has been a long threat against the US. What we're finding is that these attack groups are using Western technology now to help further scale and automate their attacks against western systems. I think that's that's the part where it's like, Okay, now we have solid proof other nation states are actually doing this against us.

But the China angle is only one troubling aspect of the report. Anthropic described the cyberattacks as being almost fully autonomous. AI carried out up to 90% of the cyberattacks. Each hacking campaign only needed about six human prompts. Anthropic detected the attacks in September and notified the affected companies. They also hardened their own security measures.

Artificial Intelligence may be sophisticated, but it isn’t at the level of carrying out attacks without human prompts along the way. Amar Singh is CEO of security consulting firm Cyber Management Alliance, gives an example of what AI can’t do, at least not yet:

SINGH: Go and hack this bank and withdraw 1 million full stop and AI will do everything that it takes to think and rethink and rethink and rethink until it succeeds and withdraws 1 million and comes back to you and gives you that million. I think the good news is we're not there yet.

He adds that humans will always be needed to give the initial hacking command. But Victor Benjamin believes it won’t be long before AI can do the rest of the work, without prompts.

BENJAMIN: So the whole concept of prompt engineering will probably go away in the near future as LLMs, you know, you won't have to manipulate the LLM to give you the output you're seeking for. It's going to better understand the nature of your request. So so it will require less prompting in the future. As far as being completely autonomous, I don't, I don't see why not.

More intuitive cyberattacks are concerning because the world is becoming what Benjamin calls “cyberphysical.” Computers have become integrated into just about everything from cars to dishwashers. Meaning everyday tools are suddenly vulnerable to cyberattacks.

BENJAMIN: The security issues aren’t just attacking a server and like stealing your credit card. Now they can attack autonomous vehicles or traffic lights and cause chaos in the real world, right?

Benjamin thinks it’s too late to scale back reliance on AI. Instead, he recommends fighting fire with fire.

BENJAMIN: Just like AI is being used to attack systems, we can have AI based defenders. So I think at some point, cybersecurity will become less managed by humans and more managed just directly by AI. You know, it's always going to be this cat and mouse game. That's what cybersecurity has been for decades. There's always new defenses. The attackers circumvent them and then back and forth. AI is kind of just a new frontier of it.

For WORLD, I’m Paul Butler. Bekah Bernhardt wrote and reported this story.

WORLD Radio transcripts are created on a rush deadline. 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 WORLD Radio programming is the audio record.