SentinelOne autonomous detection blocks trojaned LiteLLM triggered by Claude Code

SentinelOne reports it automatically blocked a supply-chain attack involving trojanized LiteLLM packages that executed obfuscated Python code and an infostealer, established persistence via a systemd user service (~/.config/systemd/user/sysmon.service and ~/.config/sysmon/sysmon.py), contacted a C2 server on delayed intervals to evade sandboxing, and expanded by creating privileged Kubernetes pods to deploy backdoors; the campaign was delivered via compromised maintainer credentials (from tools like Trivy) and was even triggered by an AI coding assistant (Claude Code), highlighting risks of AI agents with system-level access.