The Cloud Security Alliance (CSA) AI Safety Initiative published a research note on 7 November 2026 describing ThisBash, a reported design weakness in Anthropic’s Claude Dispatch. The note says a browser extension able to script a logged-in claude.ai page could use the account’s authenticated session to instruct a paired Claude Desktop agent, potentially on another computer. CSA explicitly says it did not independently reproduce the attack.[1]
According to the note’s account of researcher Gal Weizman’s demonstration, a script located the desktop agent’s session, submitted instructions and automatically clicked approval prompts. The agent then installed an npm package whose preinstall hook executed a payload that opened the calculator application. The reported chain used the user’s authenticated command channel directly; it did not require malicious content to be mistaken for instructions.[1]
CSA reports that Anthropic accepted the submission as valid but closed it as a duplicate, and that Dispatch remained available to existing users in late October despite a planned retirement. The note identifies no documented issue-specific vendor fix or fixed version, and reports no known in-the-wild exploitation as of its writing. Claims that similar exposure persists in Claude Cowork and Claude Code Remote Control remain unverified in the note.[1]
[1]: Cloud Security Alliance AI Safety Initiative, “ThisBash: Browser Sessions as Remote Shells via Claude Dispatch,” 7 November 2026. https://labs.cloudsecurityalliance.org/research/csa-research-note-claude-dispatch-thisbash-20261107-csa-styl/
Sources
Check or correct this page
Verification record
The digest below shows whether the text you are reading matches the published source. It does not show that the reporting is correct. The build computes it from this page's Markdown source and lists it in the site manifest.
52f5bbf06d1ba15cb7d326ec0f97702016ba4e12b3adf533e1cf85236003d7b1