Stay informed with free updates
Simply sign up to the Artificial intelligence myFT Digest — delivered directly to your inbox.
Illicit access to AI models and computing power is fast becoming the hottest commodity in the cyber criminal underworld, as hackers seek to harness expensive large language models for extortion, warfare and espionage.
John Hultquist, chief analyst for Google Threat Intelligence Group, said the cyber security unit had observed a major increase in so-called LLM-jacking this year, including the sale of stolen login credentials for public AI tools and the theft of computing resources by groups that want to run their own models for free.
“What we are seeing in the underground is a growing economy associated with access to AI,” Hultquist, a 20-year cyber security veteran, told the FT.
Cheap access to expensive AI gives cyber attackers a financial edge over their targets, who need to use the same AI tools to defend themselves, Hultquist warned.
“The bottom line is all this behaviour gives them a sort of economic or efficiency advantage against us, because they’re going to essentially be able to get their hands on these tokens at a much cheaper rate,” he said.
Marketplaces on the dark web are selling access to AI models from the likes of Anthropic, Google and OpenAI at discounts of up to 97 per cent, Google Threat’s researchers have found. AI subscriptions for the most advanced versions of ChatGPT and Claude can cost as much as $200 per user every month.
Given AI labs monitor for signs of such abuse, some of the sellers even offer “guaranteed access” by committing to provide new credentials at no extra cost if the initial account gets blocked, Hultquist said.
In other cases, criminal and state-backed groups are hacking companies’ cloud-hosted servers and dropping in their own AI models to run on the target’s system, in the same way threat actors have sought to compromise third-party machines to mine cryptocurrency. One such incident involved a “very active” Chinese cyber espionage group, which has previously targeted the US, he said.
Hultquist has been tracking state-backed hacking groups for two decades. In 2014 he exposed the activities of a prolific Russian actor that he dubbed Sandworm, linking it two years later to an unprecedented attack on Ukraine’s power grid.
AI is already being used by “every threat actor”, Hultquist said, adding that AI tools will therefore need to be integral to the future of cyber security. In Anthropic’s most recent quarterly misuse of AI report, it identified threat actors trying to use its Claude tool for malicious activity in more than two dozen countries including the US, UK and Yemen.
“Anybody who decides that AI is a fad and wants to let it just wash over them is going to wake up one day underwater,” he said. “They’re going to have more incidents, more alerts, more attacks than they’ve ever seen before. We have to get our house in order now.”
As more large corporations seek to host customised AI models on their own servers instead of renting capacity from cloud providers, those systems will also become a target and will have to be carefully secured, Hultquist warned. “If you’re paying for the compute, which can be very expensive . . . that becomes a major potential resource for a threat actor.”
Hultquist added that the best time for hackers to “sneak in” to targets’ accounts and computer servers was now, given that companies are still learning how much AI they will use. “You might think a sudden major increase in compute usage is completely normal because you have just adopted all this AI infrastructure,” he said. “It’s a real opportunity for somebody to hide in the noise.”