Stolen LLM keys run up $100,000 a day in victims' bills

Criminals are stealing corporate large language model API keys and cloud credentials to run their own high-volume AI workloads, leaving victim companies with fraudulent charges that can exceed $100,000 a day, according to a report published by Google's Mandiant incident responders on 2 November 2026.

The Register reported that Mandiant saw cumulative charges at some organisations reach $1 million before the victims revoked the intruders' access. The tactic is known as LLMJacking, and Google said it has been increasing steadily over the past two years.

In one case detailed in the report, a group Google tracks as UNC6993 gained access to LLM API keys, then generated additional keys and enabled services inside the victim's cloud projects, running up $100,000 in charges. In a separate mid-2026 incident that Mandiant responded to, the same group compromised a cloud instance through a misconfigured service account key and used that access to grant itself further access to a model API. According to the report, the group also set up a session hijacking page that posed as an application portal for joining a group of LLM users. The page tricked users into visiting the legitimate LLM service website and pasting a raw JSON payload containing an active access token, which was then posted to a backend API to bypass password and multi-factor authentication controls.

Google also documented a different thief who subscribed victim accounts to Pay-As-You-Go agreements so it could reach advanced models at the victim company's expense. That actor selected premium foundation models, launched high-volume inference streams, and used what the report calls a "burn and rotate strategy" to get around rate limits, creating and deleting short-lived service accounts while checking the billing interface to watch the financial damage it was causing.

In another incident, an intruder made more than 500,000 LLM API calls in under eight hours using stolen third-party API keys to power a token routing service called Jet Night, according to a separate Google report published the same day. Google Threat Intelligence Group said Jet Night's per-token prices are so far below typical market rates that it believes the service relies on stolen keys to stay profitable. The report said Jet Night offers an unlimited token plan for roughly $60 a month with alleged access to frontier models, against a consumer service level that would usually cost between $100 and $250 a month. The service has admitted to multiple outages on its public Telegram channel, which Google's threat hunters said is likely due to stolen keys being revoked as victims discover the compromises.

Google said a division of labour is emerging in the LLMJacking ecosystem, between criminals who establish initial access to compromised cloud and AI services and those who run the downstream monetisation infrastructure, which it calls "LLM Mixers". The company predicts criminals will increasingly prioritise LLMJacking over other forms of resource theft such as cryptojacking.

The chief analyst at Google Threat Intelligence Group told The Register that attackers want scale advantage over defenders, and achieve it with agentic tools and by stealing tokens from unwitting third parties, adding that locking down these resources protects organisations from runaway costs and the wider community from cheap, autonomous attacks.

To reduce the risk, Google advised defenders to prioritise identity and access management and supply chain security, use a centralised secrets management platform to rotate keys and secrets, apply least-privilege access to users and service accounts, and automate secrets scanning in codebases to find and remove API keys committed to source code.

The Register noted that an August report from CrowdStrike documented an extreme case in which a token thief sent about 200,000 API requests in just two minutes.