本地 Agent 在同一目录启动即被感染,完整链路已录制可复现。
ai_agentrcepoc
🚨 Interesting AI attack for initial access discovered by Forever Security recently:
Website → Auto-downloaded .md → Agent context ingestion → Command execution.
I tried to emulate the above scenario in the video:
1⃣ Malicious Markdown lands in Downloads
2⃣ Codex launched from the same directory
3⃣ Benign prompt submitted
4⃣ Hidden instructions processed
5⃣ calc.exe executed