AEGLIX News
Google has alerted that malicious actors are commandeering corporate access to Large Language Models in order to fraudulently exploit costly compute capacity and AI services, a scheme dubbed LLMjacking. While no proof exists that the platform is misusing its standing to collect employee prompts or manipulate outputs, the research underscores the security dangers tied to deploying these tools.
Victims of LLMjacking span North America, South America, and the Asia Pacific region, covering the financial services, retail, technology, and transportation industries. In one day, illicit usage generated more than $100,000 USD in charges, and some organizations saw cumulative bills climb to $1 million USD before access was cut off. The primary entry point has been compromised or exposed cloud identity tokens such as IAM credentials, Service Account keys, GitHub Personal Access Tokens, and active session tokens.
Threat actor UNC6993 infiltrated a cloud environment via a misconfigured Service Account key and leveraged it to reach an LLM API. The group also exploited CVE-2026-40217 in LiteLLM to elevate privileges. UNC6993 then generated additional API keys to expand operations inside victim cloud projects, accruing over $100,000 USD in billing fees. Separately, LLM Mixers operate as covert routing hubs that proxy, aggregate, and resell unauthorized LLM access to other criminals through underground marketplaces.
A grey market relay service deploys authentic OpenAI Codex and Anthropic Claude Code CLIs onto enterprise machines and reconfigures them so every request travels through third-party infrastructure rather than the vendors' own servers. This relay operation is assessed to have been active since at least early 2026, chiefly affecting around a dozen organizations in Mainland China as well as multinational subsidiaries in South Korea, Taiwan, and Malaysia.
Report prepared 2026-11-05 · Based on reporting from The Hacker News. Source publication date above; subsequent developments may not be reflected.