"LLM-jacking" is becoming industrialized: access to paid models resold at up to 97% less

Credentials stolen from AI services are sold on underground marketplaces, sometimes with guaranteed replacement if the account is blocked.

The diversion of artificial intelligence subscriptions became a criminal activity in its own right in 2026. The principle: to retrieve AI service identifiers, or directly enter computing infrastructures, to run expensive models on someone else's bill. The practice inherited the name "LLM-jacking".

John Hultquist, Chief Analyst of the Google Threat Intelligence Group, describes a sharp rise in attacks on AI accounts and computing resources during the year. In clandestine marketplaces, access to OpenAI, Anthropic or Google models are offered with discounts of up to 97% of the official price; Some vendors even promise to replace IDs if the account is detected and disabled.

The collection is on a large scale: a campaign called Bissa Scanner has harvested IDs from AI suppliers in over 30,000 configuration files left exposed. Beyond the simple free calculation, these stolen accesses are also of interest to actors who see them as a tool for extortion, intelligence or military operations.