Vulnerability / research
Claude Dispatch research demonstrates unintended commands on a connected desktop
Reported
Published here
What happened
Forever Security demonstrated that a browser extension with access to a signed-in Claude session could use Dispatch to make the connected Claude Desktop application execute commands. The desktop could be on another device using the same account.
Parties and reporting
AI tools and systems: Claude Dispatch, Claude Desktop, Authorized browser extension
Outcome
The researcher says Anthropic acknowledged the issue in May as a duplicate of an internal finding and later indicated plans to retire Dispatch. Anthropic’s Help Center says existing users retain access. A completed fix is unverified.
Operating lesson
Review which extensions can change content in agent web sessions and which accounts can dispatch work to managed desktops.
Evidence and limits
November 4 is the public byline. The May 14 submission and May 19 acknowledgement were private. The demonstration required extension access to the signed-in site and a connected desktop. The sources establish no malicious use.
Sources
- Forever Security public disclosure and vendor-response timeline
Primary source ·