Malicious Browser Extension Could Use Claude Dispatch to Run Code on User's Computer | Security
A design flaw in Anthropic's Claude Dispatch, publicly disclosed in November 2026, allows a malicious browser extension with access to claude.ai — in an authen…
Published on MyPrivateClaw
Nov 6, 2026, 8:40 PM UTC
Coverage date
Nov 6, 2026
Last updated
Nov 6, 2026, 8:40 PM UTC
News summary
Claude Desktop is an Electron application built largely around the same web interface and API as claude.ai. The Dispatch interface and API were reachable from an ordinary authenticated web session, and the service accepted Dispatch requests from the browser even though the web version simply didn't show a link to Dispatch. The shared web/desktop architecture is the precondition that leaves Dispatch reachable from the browser without a separate authorization signal. In the proof of concept, a MAIN world content script registered on claude.ai called GET /v1/sessions?tags=cowork dispatch local to retrieve the desktop agent's session ID, then used prompt forcing and auto clicked approval buttons to get Claude to install a malicious npm package and run a command on the operating system. The PoC payload opened Calculator to demonstrate local code execution; the researchers characterize a real…