Solutions  ·  2026-06-21

Microsoft AutoJack: RCE Exploit Chain in AI Agent Frameworks via MCP WebSocket

SolutionsHigh impactGlobal
Microsoft Defender Security Research Team disclosed AutoJack on June 18, 2026 — a three-vulnerability exploit chain in AutoGen Studio (Microsoft Research's agentic prototyping UI) allowing untrusted web content rendered by a browsing agent to reach local MCP WebSocket and spawn arbitrary processes on the host. Attack crosses localhost trust boundary without authentication.
Demonstrates systemic risk in agentic AI frameworks: agents with web-browsing and privileged local-tool access create RCE surface. Highlights that localhost is no longer a trust boundary once agents access both open web and local services. Affects developers using AutoGen Studio in production-like environments.
Developers and ML engineers using AutoGen Studio; organizations deploying browsing agents with local tool integration. Immediate remediation guidance: isolate agents from privileged local services.
Sources
Microsoft Security Blog: AutoJackThe Hacker News: AutoJack Attack
See this in the live feed Explore related AI security and governance findings — updated every morning.
Open the feed →