LLMJacking: Stolen AI Access Is Racking Up Bills of Up to $1m for Victims

Google's Mandiant business has warned of a growing threat to company AI systems called "LLMJacking", where cybercriminals steal access to large language models (LLMs) and sell it on. Mandiant said it has responded to multiple incidents this year across financial services, retail, technology and transportation.

Unlike cryptojacking, which tends to drain resources gradually, LLMJacking can produce large spikes in fraudulent cloud charges within hours. Mandiant observed charges above $100,000 in a single day, and at some organisations the total reached $1m before access was revoked. Attackers bypass public-facing chatbots and go after backend model resources, typically using LLM API keys or exposed or stolen cloud identity tokens. These include IAM credentials, Service Account keys, GitHub Personal Access Tokens and active session tokens. Once inside, they escalate privileges and look around for more credentials. They may also target vulnerabilities in LLM routing software.

Google has seen financially motivated groups using these methods across North America, South America and Asia Pacific. It also sees a division of labour: some actors focus on gaining access, while others, known as "LLM Mixers", run gateways that proxy, aggregate and resell access to criminal customers. Because the work is so lucrative, Google expects criminals to prioritise LLMJacking over cryptojacking going forward.