Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

NewsResearch

Google Threat Intelligence: LLMjacking crews turn stolen cloud credentials into resold AI access, with charges reaching 1 million dollars at some victims

Google Threat Intelligence Group and Mandiant described on November 2 how attackers make money from stolen cloud and model access. Exposed or stolen cloud identity tokens are the main way in: IAM credentials, service account keys, GitHub personal access tokens and live session tokens, alongside LLM API keys taken from public repositories and exposed model servers such as Ollama. Keys are tested with max_tokens=1 calls to stay under volume alerts. Once inside, attackers enable managed AI services, create API keys, raise GPU quotas and use short-lived service accounts to get past rate limits, then resell the access through "LLM mixers" built on open-source gateways. One service, Jet Night, sells an unlimited plan for about 60 dollars a month and is believed to rely on stolen keys; one intrusion tied to it made more than 500,000 calls in under eight hours and cost a victim nearly a million dollars in tokens. Google saw charges above 100,000 dollars in a single day and up to 1 million dollars at some organizations before access was cut, in financial services, retail, technology and transportation across North America, South America and Asia Pacific. It also names a LiteLLM flaw, CVE-2026-40217, among the ways in.

Why it matters

A leaked cloud token is now a direct bill for someone else's AI use. Deny AI services that a project does not need, cap spend and quotas per key, and alert when a model service is turned on outside your deployment pipeline.

Sources

Read it with the rest of the news in the Learning Center