IT Home, September 28 — According to the Financial Times on the 26th, illegally obtaining AI models and computing power is rapidly becoming the most sought-after commodity on the cybercrime black market, as hackers aim to leverage expensive large models for ransomware, cyber warfare, and espionage.
John Hultquist, Chief Analyst at Google's Threat Intelligence Team, disclosed that so-called "LLM hijacking" activities have significantly increased this year, manifesting in the sale of stolen login credentials for publicly available AI tools and some groups hijacking others' computing power to run their own models for free.
Holtquist, with 20 years of experience in cybersecurity, admitted: "We are seeing an economic system around AI usage rights grow steadily in underground markets."
Holtquist warned that attackers can leverage expensive AI tools at a low cost, gaining a cost advantage, while defenders must also use these same AI tools to protect themselves. "Ultimately, these practices give them some economic or efficiency advantage in their confrontation with us, because they can acquire these tokens at a much lower price."
Researchers from Google's Threat Intelligence Team have found that access to AI models from companies such as Anthropic, Google, and OpenAI is being sold on dark web markets, with discounts of up to 97%. For comparison, the highest-tier subscriptions for ChatGPT and Claude can cost up to $200 per user per month (IT Home note: approximately RMB 1,346 at current exchange rates).
Holtquist said AI labs monitor for signs of abuse, so some sellers have even launched "guaranteed access" services, promising to provide new login credentials at no extra cost if the initial account is banned.
Some criminal groups and state-sponsored organizations infiltrate servers hosted in the cloud by businesses and implant their own AI models to run them using the victims’ systems. This technique is similar to how threat actors attempt to compromise third-party devices and borrow their computing power for cryptocurrency mining.
Holtquist warned that an increasing number of large enterprises want to deploy custom AI models on their own servers rather than renting computing power from cloud providers, making their internally operated AI systems prime targets for attacks that must be rigorously protected. “If you have to pay for the computing power yourself—and the costs can be very high—that computing power becomes a valuable potential resource for threat actors.”
Holtquist believes that companies are still figuring out how much AI they actually need, making this an ideal time for hackers to quietly infiltrate target accounts and computer servers. "You might think a sudden spike in computing power usage is completely normal because you've just adopted these AI infrastructure tools. And that’s exactly the opportunity hiding in the noise."