# LLMJacking Attacks Cost Organizations Over $100,000 Daily in Illicit Token Charges

Criminals are hijacking corporate LLM access to exploit expensive cloud computing resources for profit.

- **Source:** [www.theregister.com - Articles](https://www.theregister.com)
- **Published:** 2026-11-02
- **Topic:** [Software & Engineering](https://brifing.si/c/engineering)
- **Original article:** <https://www.theregister.com/security/2026/11/02/token-thieves-costing-companies-100k-per-day-in-illicit-llm-charges/5306967>
- **On Brifing:** https://brifing.si/s/c4684a86-dd3d-4bba-8fbd-57030c404eec

## The Rise of LLMJacking

Criminals are increasingly hijacking corporate LLM access to steal expensive computing resources. This tactic, known as LLMJacking, has seen steady growth over the past two years. Victim organizations report daily fraudulent charges exceeding $100,000, with some cumulative losses reaching $1 million before security teams successfully revoke the intruders' unauthorized access.

## Exploiting API Key Vulnerabilities

Threat actors like the group tracked as UNC6993 gain access to LLM API keys through misconfigured cloud service accounts. Once inside, they generate additional keys and enable premium services within the victim's project. These attackers often subscribe compromised accounts to expensive pay-as-you-go agreements to maximize their illicit resource consumption.

## Burn and Rotate Strategy

To bypass rate limits, attackers employ a burn and rotate strategy by creating and deleting short-lived service accounts. This method allows them to maintain high-volume inference streams while monitoring financial impacts through the billing interface. In one extreme instance, a thief executed 200,000 API requests in just two minutes.

## The Jet Night Operation

Google researchers identified a token routing service called Jet Night that operates profitably by using stolen keys. By offering unlimited token plans for $60 per month, they undercut market rates significantly. Frequent service outages on their Telegram channel suggest that victims are actively discovering and revoking the compromised keys used.

## Specialized Criminal Labor Models

The LLMJacking ecosystem now features a clear division of labor between different criminal roles. Initial access brokers compromise cloud instances, while downstream operators manage monetization infrastructure known as LLM Mixers. This professionalization allows attackers to scale their operations efficiently, moving beyond simple cryptojacking toward more lucrative AI resource theft.

## Session Hijacking and MFA Bypass

Attackers use sophisticated session hijacking pages to trick users into revealing active access tokens. By capturing raw JSON payloads, criminals bypass password and multifactor authentication controls entirely. This technique allows them to interact directly with backend APIs, effectively masquerading as legitimate users to perform unauthorized, high-volume model inference tasks.

## Prioritizing Identity and Access

Defenders must prioritize identity and access management alongside robust supply chain security to mitigate these risks. Implementing a centralized secrets management platform is essential for rotating keys regularly. Organizations should strictly apply least-access privileges, ensuring that every service account possesses only the permissions necessary to complete its specific, defined operational role.

## Automated Defense and Monitoring

Security teams should automate secrets scanning within their codebases to identify and remove exposed API keys. This proactive approach prevents accidental leaks that facilitate unauthorized access. As LLMJacking becomes a primary target for resource theft, constant vigilance over billing interfaces and API usage patterns remains critical for protecting corporate financial assets.

## Sources covering this story

- https://www.theregister.com/security/2026/11/02/token-thieves-costing-companies-100k-per-day-in-illicit-llm-charges/5306967 — 

---

_Summary written by Brifing from the original reporting. Copyright in the original article belongs to its publisher; read the full piece at the link above._
