Google warns that hackers are hijacking enterprise AI access and running up million-dollar cloud tabs

Cybercriminals are increasingly stealing artificial intelligence credentials to hijack corporate models, running up cloud bills that exceed $100,000 in a single day and reach up to $1,000,000 cumulatively. According to researchers at Google's Threat Intelligence Group and Mandiant, these attacks—known as LLMjacking—target exposed endpoints and valid application programming interface keys to siphon expensive computing power. [1][2][3]

The scheme relies on a division of labor between initial-access brokers and downstream operators known as LLM mixers. These mixers pool stolen access using proxy infrastructure and resell cheap subscriptions, such as Jet Night's $60/month unlimited plan, to underground users. Attackers obtain credentials through compromised cloud accounts, GitHub tokens, and flaws such as CVE-2026-40217 in LiteLLM, at times driving traffic past 1,000 requests per second. [1][2]

Google warns that threat actors are likely to prioritize LLMjacking over older resource-theft schemes like cryptojacking because model usage can generate six-figure costs in minutes. To counter the threat, researchers advise companies to enforce phishing-resistant multifactor authentication, tighten identity and access management, and set strict API spending limits. [1][2]