AutoJack: How a single page can RCE the host running your AI agent

2026-06-22T08:52:20Z70827eb7b91e168b9ba1cf239142fc69ebbaf8f297bb472c3c31ff607cadfd8d
agent-securityai-agent-securityassertautogen-studioautojackbackdoorclipboard-theftcrypto-clipperforensicslocalhost-trustmastramcp-websocketmdashmitigationnpm-supply-chainpostinstall-payloadrcesupply-chain-compromisethreat-inteltor-c2unsafe-parameter-handlingworm-like-propagation

What happened

The feed highlights several active security issues and research: AutoJack — a novel exploit chain where a single malicious webpage can cause remote code execution on the host running an AI browsing agent by abusing localhost trust, missing authentication, and unsafe parameter handling (specifically targeting AutoGen Studio’s MCP WebSocket), illustrating that agent browsing + local service access breaks traditional localhost trust boundaries. A separate supply‑chain incident (Mastra) shows a poisoned npm package that delivered a postinstall payload and infected 140+ projects. Microsoft also dis

Why it matters

A reviewed impact interpretation has not been published for this record.

Evidence and limitations

Source ID
microsoft_security_blog
Record identifier
70827eb7b91e168b9ba1cf239142fc69ebbaf8f297bb472c3c31ff607cadfd8d
Enrichment time
2026-06-22T08:52:20Z
AI-assisted enrichment
Yes

This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.