CERT-EU demos AI agent plugin backdoor that steals credentials on "lgtm, ship it"
TechNadu · x · 2026-11-03
CERT-EU demonstrated a supply-chain backdoor in AI coding agent plugins: triggered by simple phrases like "lgtm, ship it" embedded in user inputs or logs, a compromised plugin silently exfiltrates environment variables, system details, and developer credentials — without generating any model tool call or altering conversation logs.
Key points:
- The risk sits at the extension/plugin layer, not the model, bypassing model-level guardrails and file-reading restrictions by executing malicious tasks at the process level
- Mitigation: treat the broader AI agent ecosystem as a security boundary — container isolation, credential rotation, and strict monitoring of external plugin executions
More from coding & agent
- Live workshop: instrument your AI agent with OpenTelemetry and debug it via Langfuse traces — Al_Grigor · 2026-11-03
- AutoRLBench: A Benchmark Testing AI Agents on Real RL Training Failures — wzenus · 2026-11-03
- Building live TV fact-checking agents on Bedrock AgentCore at re:Invent IND421 — FlolightC · 2026-11-03
- New Tutorial: Driving Gemini for Computer Use with Cua Driver — osanseviero · 2026-11-03
- SecureInput: open-source local secret input for MCP coding agents — rishabhgusain · 2026-11-03
- Sidelight runs browser agents in your real Chrome session to dodge login walls — Equivalent_Type6726 · 2026-11-03