All news

26 September 2026

Stolen AI is the new hot item on the dark web, Google's threat team says

Google's cyber security researchers say criminals are selling hijacked access to AI models at deep discounts and secretly running their own models on hacked company servers.

Criminals have found a new thing to steal: artificial intelligence. Google's cyber security unit says access to AI models and the computing power behind them has become one of the most wanted goods in the online underworld, a trend it calls LLM-jacking. An LLM, or large language model, is the kind of AI behind chatbots like ChatGPT and Claude.

What happened

John Hultquist, chief analyst at the Google Threat Intelligence Group, told the FT his team has seen a big jump in this activity this year. It comes in two main flavors.

  • Stolen accounts. Dark web marketplaces are selling logins for AI tools from Anthropic, Google and OpenAI at discounts of up to 97%. The most advanced ChatGPT and Claude plans can cost up to $200 per user a month. Because AI companies watch for abuse, some sellers even promise free replacement logins if an account gets blocked.
  • Stolen computing power. Criminal and state-backed hackers break into companies' cloud servers and quietly run their own AI models there, much like earlier hackers hijacked other people's machines to mine cryptocurrency. One case involved a very active Chinese espionage group that has targeted the US before.

Hultquist said the cheap access gives attackers a cost edge over the companies defending themselves, which pay full price for the same tools. Anthropic's latest quarterly report on misuse of its AI found bad actors trying to use Claude in more than two dozen countries, including the US.

Why it matters

For investors, this is a new line item in the AI bill. Companies are spending heavily to run AI, and more big firms want to host custom models on their own servers rather than rent from cloud providers. Hultquist warned those systems will become targets too, because the computing power inside them is expensive and therefore valuable to steal.

There is also a timing problem. Many companies are still working out how much AI they will use, so a sudden jump in computing use may look normal rather than suspicious. Hultquist said that makes now a good moment for hackers to "hide in the noise." In other words, the AI boom is also a boom for people who want to borrow it without paying.

What to watch

  • Whether AI companies tighten account checks, which could add friction for paying users.
  • Cyber security spending plans as companies roll out their own AI servers.
  • More public reports from AI labs on how their tools are being misused.

Based on reporting by the Financial Times: Hackers hijack AI accounts and servers to fuel new cyber crime boom. Facts, not advice.

More news