Image: pynt.io
Code execution through email: using Claude to hack itself
A security write-up showing how a crafted email could make Claude run code.
Credits
- HN submitter nonvibecoding Source ↗
Earliest dated source:
The full story
Golan Yosef, a security researcher at Pynt, showed how a crafted Gmail message could lead to code execution through Claude Desktop when it's connected to both an email MCP server and a tool that runs code. Claude itself helped plan the attack. His point: in AI systems the danger often isn't one vulnerable app but the combination of untrusted input and powerful tools. The Hacker News discussion (137 points) echoes that connecting a code runner to untrusted input is asking for trouble.
How it was done
Tools
Themes
Loading themes…
Sources
From the maker's own write-up or code. We summarise what they report; we haven't tested it ourselves.
- Hacker News: Code execution through email: How I used Claude to hack itself
The full write-up of the attack chain.
- Hacker News discussion (137 points)HN submitter: nonvibecodingPosted
Hacker News discussion (137 points).