New Developments in LLM Hijacking Activity

**Executive Summary:** Wiz Research observed a campaign by JINX-2401 abusing compromised AWS IAM credentials to attempt Bedrock model hijacking across multiple environments; the actor created IAM users matching the regex ^[A-Z][a-z]{5}[0-9]{3}$ and a policy named New_Policy granting bedrock invocation permissions, used Proton VPN IPs and Python/Boto3 user-agents, and attempted model agreement/API calls that were mostly blocked by Service Control Policies (SCPs); the report supplies IOCs, detection queries, and SCP-based mitigation recommendations.