What happened

  • The report titled “The Rise of LLMJacking: How Threat Actors Monetize Compromised Cloud and Model Resources” documents multiple cases where attackers used stolen cloud credentials or exposed LLM endpoints to run up fraudulent charges.
  • Victims in North America, South America and Asia-Pacific, including financial services, retail, technology and transportation firms, saw daily bills exceed $100,000, with one organisation reaching nearly $1 million in eight hours.
  • Attackers obtained access through compromised service-account keys, GitHub personal access tokens, or open-source LLM servers such as Ollama. A group tracked as UNC6993 broke into a cloud instance through a mis-configured service account key and hosted a fake sign-up portal that harvested real API tokens.
  • Some actors scraped API keys from public repositories and sent tiny requests that evaded volume-based alerts. Others exploited CVE-2026-40217 in LiteLLM’s key regeneration endpoint to turn ordinary virtual keys into master administrator credentials.
  • Stolen credentials were either burned directly, generating massive usage on services like Amazon Bedrock and Google Gemini, or sold through “LLM Mixers” such as the Jet Night token routing service, which offered unlimited model access for about $60 a month.

Why it matters

  • The speed of cost escalation far outpaces traditional cryptojacking, which drains resources gradually. LLMjacking can spike bills within hours, catching organisations off-guard.
  • Exposed API tokens and mis-configured endpoints become high-value targets as enterprises adopt higher-tier models and agentic workflows, expanding the attack surface.
  • The report notes many victims had never enabled LLM services, yet attackers turned them on using stolen keys, showing how a single compromised credential can open a cascade of services.
  • Financial impact is severe: a single compromised account can generate $100,000+ in charges, and the cumulative effect can reach millions, threatening both operational budgets and reputations.
  • GTI Group expects LLMjacking to overtake cryptojacking because mixers provide quick, high-margin returns for attackers.

What this means for enterprise IT teams

  • Any unauthorised enablement of LLM services should be treated as a critical security event. Teams must monitor for unexpected activation of services such as Amazon Bedrock or Google Gemini.
  • Implement spending caps and scope API keys to specific models and environments to limit exposure.
  • Enhance log monitoring to detect single-token pings that validate stolen keys, a technique used to stay under volume alerts.
  • Rotate and tightly control service-account keys and GitHub tokens; avoid exposing them in public repositories.
  • Conduct regular audits of open LLM endpoints and enforce network restrictions to prevent internet-facing exposure.
  • Consider adopting automated credential-scanning tools that can flag mis-configurations before they are exploited.

In the words of John Hultquist, chief analyst, Google Threat Intelligence Group: "Locking down these resources will not only protect organizations from runaway costs,"

The bottom line

LLMjacking represents a new, high-impact vector for cloud abuse, capable of inflating bills by six figures in a matter of hours. Enterprises must tighten credential hygiene, enforce strict spending controls, and monitor for unauthorised LLM service activation to defend against this rapidly evolving threat.

Sources

  • siliconangle.com