Skip to content
Bamboo Flags
Custom automation & AI for growing businessesDiscover · Design · Deliver · Drive

Stolen AI API keys: LLMjacking bills reach $100,000 a day

Google reports attackers who steal AI API keys and cloud logins, then resell the access. Bills passed $100,000 in one day. What happened and the checks to do.

Bamboo Flags3 min read

On 2 November 2026, Google Threat Intelligence Group and Mandiant published a report on "LLMjacking". In LLMjacking, an attacker steals a company's access to paid AI models, then uses or sells it. The company gets the bill. In the incidents that Mandiant investigated, charges passed $100,000 in a single day. At some organizations, the total reached $1 million before the access was stopped.

This applies to any business with an API key for OpenAI, Anthropic, Google Gemini or a cloud AI service, also when the key sits in a script, a no-code tool or an automation platform.

How attackers get the keys

The report names these entry points, from incidents in 2026 in financial services, retail, technology and transportation:

  • Stolen cloud credentials, the most common one: cloud logins, service account keys, GitHub access tokens and active session tokens.
  • AI API keys in public code. One attacker collected company API keys from public GitHub repositories.
  • AI servers open to the internet. One attacker added its own models to open Ollama servers.
  • A fake login page that looked like an application portal. Users pasted an access token into it, and the attacker used the token without a password or a second factor.

In several incidents, the company did not use the AI service at all. The attacker used a stolen cloud login to turn on Amazon Bedrock or the Gemini API and ran up the bill in hours.

Stolen access is resold at a discount

Some groups steal the access. Other groups resell it. Google calls these resellers "LLM Mixers".

One of them, Jet Night, made more than 500,000 AI API calls in less than eight hours with stolen keys. It sold an unlimited plan for about $60 a month, while comparable plans cost $100 to $250. The Register reports that Jet Night announced outages on Telegram, which Google links to victims who found and revoked their keys.

Some attackers first test a key with requests for one token of output. In the report, these tests were too small to set off volume alerts.

Trigger.dev customer environment variables were exposed

Trigger.dev is a hosted platform for AI agents and workflows. On 1 November 2026 it posted a security disclosure: an attacker got access to the environment variables of "a subset of organizations". Many tools keep their API keys and passwords in environment variables. Trigger.dev says that it blocked the exploit and emailed the steps to fix the problem to the owners of the affected accounts.

API keys were also the target in the Cursor extension backdoor last week, which sent the names of key and token variables to the attacker.

What is not known

  • Trigger.dev did not say which variables were read, how many organizations are affected or how the attacker got in. It told The Register to "wait for a full incident report."
  • No source connects the Trigger.dev incident to the LLMjacking groups in the Google report.

A spending limit stops the bill, an alert does not

Google's advice includes "spending caps to set budgets and avoid over-spending". Check what your provider's setting does. On the Anthropic API, you can set your own monthly spend limit on the Billing page, and requests stop when usage reaches it. In Google Cloud, an alerts-only budget "doesn't automatically cap" usage or spending. A spend cap budget can pause the services that support it.

Checks for your AI API keys

  1. List every AI API key that your business has and where it is stored, including automation tools, plugins and contractors.
  2. Set a spending limit at each provider that stops requests, not only an alert. Set it near your normal monthly use.
  3. Give each tool its own key. Then you can revoke one key without stopping the others.
  4. If you use Trigger.dev, look for its email about the incident, also in spam. Rotate the keys and passwords that you keep in its environment variables.
  5. Search your code repositories for API keys, and turn on secret scanning. Google recommends that you "automate secret scanning within codebases".
  6. In your cloud accounts, turn off AI services that you do not use. Set an alert for when someone turns one on.
  7. Turn on phishing-resistant two-step login, such as a passkey or a security key, for cloud and AI provider accounts.

Found something worth fixing?

Bring it to a free 45-minute call.

Book a free 45-minute call

More from the blog