Summary
- CERT EU analysed 358 open source reports for its September Cyber Brief.
- The agency highlighted increased LLM jacking involving stolen AI service access and hijacked cloud infrastructure.
- The briefing also records spyware activity and several China linked campaigns.
Cybercrime markets are increasingly selling stolen access to premium artificial intelligence services, while compromised cloud infrastructure is being used to run AI workloads at somebody else’s expense, according to CERT EU.
Its September Cyber Brief draws on 358 open source reports and highlights increased activity commonly described as LLM jacking, where criminals steal or resell access to paid AI tools and abuse compromised cloud servers for unauthorised workloads.
CERT EU attributes the underlying finding to Google threat intelligence research, which reported a surge in the activity and identified cybercrime actors monetising access to premium AI services. At least some of the observed abuse was also connected by researchers with a China linked operator using victim infrastructure.
Unlike attacks that target the security of a model directly, LLM jacking turns the services and computing capacity attached to a compromised identity into assets that can be consumed or resold. An attacker may therefore profit from an account without first stealing business information or deploying ransomware.
Cloud identities now carry additional value
As organisations attach paid AI services and substantial computing capacity to cloud accounts, traditional identity compromise can open access to resources with immediate commercial value alongside email, documents, and administrative services.
The same CERT EU review records other activity involving identity and cloud access, while its wider September assessment covers spyware directed at Serbian democracy activists and dissidents as well as several campaigns linked by researchers or authorities to Chinese actors.
Those developments are not presented as one connected operation, since the Cyber Brief brings together cybercrime, cyberespionage, vulnerabilities, policy, law enforcement, and emerging technology to give EU institutions a broader view of the threat environment.
Within that environment, LLM jacking shows AI infrastructure entering an established criminal economy built around stolen credentials and cloud access rather than creating an entirely separate form of intrusion. The entry point may remain a compromised identity even when the resource being abused is relatively new.
Unauthorised AI workloads can also create direct costs for the victim because cloud compute and commercial model access are billed services. Where criminals resell that capacity or consume it at scale, the financial effect can occur before an organisation discovers any conventional data theft.
CERT EU publishes its monthly brief under TLP:CLEAR, allowing its assessment to be shared outside the institutions and bodies it directly serves.
