Thieves run up $100K a day in AI bills on other companies’ accounts

In short: criminals increasingly skip stealing data and steal access to paid AI models instead, then use it until someone notices the bill. On 2 Nov 2026 Google’s threat intelligence group (GTIG) and Mandiant described a series of such break-ins from 2026 across North America, South America and Asia Pacific. The tactic has a name: LLMjacking.

What happened

• In individual incidents the fraudulent charges topped $100,000 in a single day, and at some organisations they reached $1 million in total before access was cut off.
• A group tracked as UNC6993 got in through a misconfigured cloud service account key, generated more keys and switched on AI services itself. The bill passed $100,000.
• The same group ran a fake company portal page where users pasted their active login token, which bypassed both password and multi-factor sign-in.
• Another attacker made more than 500,000 requests in under eight hours with stolen keys, passing nearly a million dollars in costs to the victim.
• Stolen access is resold by middlemen called an LLM Mixer. Jet Night advertises an “unlimited” plan for about $60 a month, while typical consumer plans cost $100–250. Google believes that price only works because the keys are stolen.

How to defend yourself

Google’s recommendations include:

• keep every API key in a secrets manager and rotate it automatically,
• apply least privilege, including access to AI services,
• set spending caps and separate keys for test and production,
• scan code repositories for pasted keys and remove them fast,
• treat any unplanned activation of a cloud AI service as a critical alert,
• watch for oddly tiny requests that thieves use to check whether a key still works.

How to read this

This report comes from a company that sells cloud and security tools, so some advice points to its own products. The numbers, however, come from specific incidents Mandiant handled. Google expects criminals to prefer this method over hidden crypto mining, because it turns into money faster. I write about protecting code and keys in the AI code security hub.

Sources

The Rise of LLMJacking: How Threat Actors Monetize Compromised Cloud and Model ResourcesGoogle Threat Intelligence Group i Mandiant · Nov 3, 2026 · Primary sourceToken thieves costing companies $100K+ per day in illicit LLM chargesThe Register · Nov 3, 2026