Attackers hijacking companies’ access to artificial intelligence models have generated fraudulent cloud charges exceeding $100,000 in a day, with cumulative bills at some organizations reaching $1 million, according to Google researchers. Bleeping Computer and The Hacker News reported the findings from Google’s Threat Intelligence Group and Mandiant.
Known as LLMjacking, the attacks turn compromised access into a resale business. Criminals find exposed model endpoints or obtain valid credentials, then supply access to intermediary services that route subscribers’ prompts through victims’ accounts. Cloud providers bill those requests as ordinary consumption.
Bleeping Computer reported that one attacker made more than 500,000 model API calls in under eight hours using stolen third-party keys to support the Jet Night routing service. Google attributed nearly $1 million in token costs charged to one victim organization to that service.
The Hacker News described another route into victims’ accounts: a threat group Google tracks as UNC6993 operated a deceptive page inviting people to join an LLM user group. According to Google, it persuaded users to submit data containing active access tokens, allowing attackers to bypass password and multifactor authentication checks.
The exposure can extend beyond an AI bill. Researchers described attackers searching repositories, logs and secret-management systems for more credentials, creating backdoor accounts and adding SSH keys to preserve access. The Hacker News reported affected organizations across financial services, retail, technology and transportation in the Americas and Asia Pacific.
Google recommends limiting account permissions, centrally managing and rotating secrets, and setting API usage and spending caps. It also advises monitoring cloud audit and inference logs for unusual consumption, unauthorized service activation and efforts to disable logging.
Reader comments (0)
Bring evidence, ask questions, or challenge our reporting. Every comment is checked before publication.
No approved comments yet
Be the first to ask a thoughtful question or contribute useful context. We won't invent reader conversations.
Join the discussion
Members-only commenting
Everyone can read comments. To write, verify your email and sign in to your free member account.
Sign in or join freeNo newsletter subscription is required. Your email address remains private.