LLMJacking: Criminals Are Hijacking Business AI Access and Running Up Six-Figure Bills in Hours

Mandiant responded to multiple incidents in 2026 in which attackers took over organisations' access to Large Language Models (LLMs), the technology behind AI chatbots and assistants. The tactic is known as "LLMJacking". The intrusions spanned several cloud and AI providers and affected organisations in North America, South America and Asia Pacific, across financial services, retail, technology and transportation.

The financial impact can be fast and severe. Unlike cryptojacking, which tends to drain resources gradually, LLMJacking can produce huge spikes in fraudulent charges within hours. Investigators saw charges above US$100,000 in a single day, and at some organisations the total reached US$1 million before access was shut off.

Google Threat Intelligence Group has identified a range of financially motivated groups operating globally, and a growing division of labour among them. Some actors focus on gaining initial access, while others run the infrastructure that makes money from it, known as "LLM Mixers". These act as illicit gateways that route, bundle and resell unauthorised AI access to underground subscribers. In one investigation, two separate groups worked together to compromise and monetise a company's LLM infrastructure. The full report goes on to describe the attack lifecycle and offers detection, hardening and remediation guidance.