LLM Hijacking Targeting AWS (Campaign)

Wiz Threat Research identified JINX-2401 on 2024-11-26 as a threat actor leveraging compromised IAM user keys to access multiple AWS environments and attempt hijacking of Bedrock LLM models. The attacker created privileged IAM users/policies, attempted to complete foundation model agreements via API calls (e.g., PutUseCaseForModelAccess, CreateFoundationModelAgreement) and used Proton VPN IPs, but Service Control Policies prevented successful model invocation; consistent naming patterns indicate a targeted, systematic campaign.