LLM-jacking — Hijacked AI Accounts and Cloud Resources Fueling Cybercrime

This page was written and is updated by the paper from the source articles listed below. Importance is how central the topic is to current coverage; confidence is how well-sourced the page is; mentions counts the articles that fed it. How the wiki works →

## Overview Security researchers have identified a surge in 'LLM-jacking' — attacks in which hackers hijack corporate AI accounts and cloud-hosted AI infrastructure to fuel a new wave of cybercrime (FT, September 26, 2026). The technique involves compromising credentials or API keys tied to expensive AI compute resources, allowing attackers to run unauthorized workloads, resell access, or launch further attacks at the victim organization's expense. ## Mechanics According to security researchers cited by the FT, attackers are increasingly targeting the costly AI resources that companies operate, hijacking accounts and servers to run their own operations rather than paying for compute themselves (FT, September 26). This mirrors earlier 'cryptojacking' trends but is adapted to the far higher value of GPU/AI infrastructure and API access to frontier models. ## Context and Significance The emergence of LLM-jacking coincides with broader concerns about AI infrastructure security, including OpenAI's disclosure that its own models were reportedly used by hackers to compromise 'dozens' of organisations, including governments (FT, September 25, 2026). Together these developments point to AI systems becoming both a target and a tool for cybercriminals — raising the stakes for enterprises that have rapidly scaled AI deployments without commensurate security investment. ## Why It Matters For enterprises and their counsel, LLM-jacking represents an emerging category of cyber risk with implications for: - Cloud services agreements and liability allocation for compromised AI compute - Insurance coverage for AI-resource-related losses - Regulatory disclosure obligations when AI infrastructure is hijacked to attack third parties - Vendor security diligence for companies purchasing AI infrastructure/API access This is likely to be a recurring theme as AI infrastructure spend and attack surface both grow through 2026.