Google Warns of 'LLMjacking' Attacks Costing Enterprises Up to $1 Million in Cloud Charges

Google, in collaboration with Mandiant, has issued a stark warning about a new form of resource hijacking targeting enterprise Large Language Models (LLMs), which they've termed 'LLMjacking.' This attack vector involves threat actors gaining unauthorized access to an organization's LLM accounts, primarily by stealing API keys through phishing, exposed source code, misconfigured environments, or client-side code vulnerabilities. Once access is established, attackers leverage these accounts to run their own AI models, with the victim organization footing the bill. This development is critical for cloud and DevOps professionals because it represents a significant financial risk that can manifest with alarming speed. Unlike cryptojacking, which typically involves a more gradual consumption of resources, LLMjacking can lead to fraudulent charges exceeding $100,000 in a single day, with cumulative costs reaching up to $1 million before the compromise is detected and access revoked. The impact extends across various sectors, including financial services, retail, technology, and transportation, affecting organizations in North America, South America, and Asia Pacific. This trend fits into the broader context of evolving cloud security threats, where resource abuse and credential compromise remain persistent challenges. The increasing adoption of LLMs and AI services in enterprise environments creates new attack surfaces. The monetization potential for attackers is high, as they can resell access to these powerful models through underground 'LLM Mixers,' making LLMjacking a more attractive target than traditional resource theft. This also underscores the ongoing struggle with identity and access management in complex cloud ecosystems, where a single compromised API key can lead to substantial financial and operational damage. In practice, organizations must prioritize stringent API key management, including secure storage, rotation policies, and least privilege access. Regular security audits of cloud environments and code repositories are essential to identify and remediate exposed credentials. Implementing robust monitoring and alerting for anomalous LLM usage patterns and sudden spikes in AI service consumption is crucial for early detection. Furthermore, practitioners should review and strengthen their security posture around client-side code and development environments to prevent API key leakage. The exploitation of vulnerabilities in LLM routing software, such as CVE-2026-40217 in LiteLLM, also highlights the need for diligent patching and secure configuration of AI infrastructure components. Organizations should also consider implementing AI governance frameworks that include strict controls over AI agent usage and access to prevent unauthorized or excessive resource utilization.