On November 2, 2026, Mandiant and the Google Threat Intelligence Group published “The Rise of LLMJacking,” based on 2026 incident response cases in which attackers took over organizations’ access to large language model services across several cloud and model providers. Victims were in North America, South America and Asia Pacific, in financial services, retail, technology and transportation.
The report says stolen cloud identity tokens and API keys are the main way in, often keys left in public code repositories. Attackers then switch on managed AI services, often ones the victim had not enabled, create keys, run high-volume inference and resell the access through “LLM Mixers,” gateways that pool stolen keys for underground subscribers. One reseller made more than 500,000 API calls in under eight hours, passing nearly 1 million dollars in token costs to the victim. “Unlike the gradual resource consumption associated with cryptojacking, LLMJacking can generate large spikes in fraudulent cloud charges in a matter of hours.”
Why it matters: AI model access has become a commodity worth stealing in its own right, and the bill lands fast. Google’s main advice is to treat any unauthorized enabling of an AI service as a critical security event and to cap spending per model.
What it does not show: the report gives no count of total victims or industry-wide losses, and the indicator set is available only to registered users.