SecurityNov 3, 2026, 2:04 AM UTC·2 min read

LLMjacking: Hackers Run Up $100,000-a-Day AI Bills for Firms

Mandiant and Google's threat intelligence group say LLMjacking, the theft of AI access billed to victims, has become industrialized, with 2026 incidents across four sectors.

Illustration of a hooded hacker at a laptop with glowing AI network icons and a rising cloud bill. — Image: L'Usine Digitale

SUMO

Dialo

ediviewer

Attackers are increasingly using artificial intelligence at someone else's expense. In a report published on 2 November 2026, Mandiant, Google's cybersecurity subsidiary, and the Google Threat Intelligence Group (GTIG) describe how this practice, known as "LLMjacking", has become industrialized. The headline figure: costs of more than $100,000 per day for affected organizations.

What LLMjacking means

The principle is simple: a hacker uses AI services and another company pays the bill. The term was coined in 2024 by the cybersecurity firm Sysdig, and the practice was already documented at that point. What changes with the new report is the scale and the level of organization behind it.

Incidents across several sectors

Mandiant and GTIG illustrate the trend with several incidents that took place in 2026. They hit companies in finance, technology, retail and transportation, which shows that the problem is not confined to one industry.

For businesses, the risk is as much financial as it is technical. A compromised account with access to AI models can generate usage charges without the owner's knowledge, and the invoice lands on the victim rather than on the person who made the requests. The sums cited in the report show how quickly such a bill can escalate.

Why it matters

As companies roll out generative AI tools and connect them to their cloud environments, the access credentials and accounts that unlock those services become targets in their own right. The report's central message is that LLMjacking should no longer be treated as a marginal curiosity but as an established, organized form of fraud.

The publication of the report by two of Google's security teams adds to the documentation of the phenomenon first named by Sysdig two years ago, and gives security and finance teams concrete cases from 2026 to use when assessing their own exposure.