Google Threat Intelligence Group reports a significant rise in LLM-jacking attacks this year. Hackers are stealing login credentials for public AI tools and hijacking cloud servers to run their own models without authorization. This trend mirrors past activities where attackers compromised systems for cryptocurrency mining.
The underground market for AI access is expanding rapidly. Researchers observed dark web marketplaces selling access to models from companies like Anthropic, Google, and OpenAI at discounts of up to 97 percent. Some sellers offer guarantees, promising free replacement credentials if the original accounts are suspended by the service providers.
John Hultquist, chief analyst at Google Threat Intelligence, notes that attackers gain a significant economic advantage by accessing expensive AI resources at a fraction of the cost. With premium subscriptions for services like ChatGPT and Claude reaching $200 per user per month, these stolen resources become highly attractive targets for cybercriminals.
The threat landscape is global, with Anthropic reporting that actors from over 20 countries, including the US, UK, and Yemen, have attempted to use its Claude tool for malicious purposes. As businesses increasingly deploy custom AI models on their own infrastructure, these systems are becoming primary targets for exploitation.
Hultquist warns that the current period of rapid AI adoption provides a unique window for attackers. Because companies are still establishing their baseline AI usage, sudden spikes in computing power consumption often go unnoticed, allowing hackers to operate undetected within the noise of legitimate infrastructure growth.
