Problem Solutions
A Browser Extension Can Make Claude Desktop Run Code on Your Computer: How the ThisBash Flaw Works
A browser extension running on claude.ai can make Claude Desktop run code on your computer, security researchers have shown, without exploiting the browser or the operating system. The attack, named ThisBash by its discoverers at Forever Security, uses Claude Dispatch, the feature that lets you send tasks from your phone to Claude Desktop, and turns it into a remote code execution channel. Coverage by The Hacker News on November 6 and The Register on November 5 describes a flaw Anthropic has known about since May and has chosen to retire rather than patch.
What Dispatch is
Dispatch lets you send a task from your phone to Claude Desktop, where the agent works with local files, connectors, plugins and applications on the computer. It is a bridge between the Claude you use in the browser and the Claude agent that can act on your machine. That bridge is the problem: Dispatch's interface and API also accepted requests from an ordinary logged-in session on claude.ai, with no visible link or button on the website suggesting it was reachable from the page.
How the attack works
Forever Security's demonstration ran in six steps, in plain terms. A browser extension that can run scripts on claude.ai injects a script into the page, sharing the page's JavaScript context and session cookies. The script opens Dispatch in a small 200 by 200 unfocused popup window, reads the session identifier of the user's Claude Desktop instance, and then sends four prompts as if the user had written them. Those prompts led Claude to install a package from the npm registry that the researchers had prepared, and the package's install script opened the Calculator app, a visible sign that local code execution had been achieved.
The researchers say the same step could have been used to steal files or install a backdoor. The demonstration did not carry out those actions. The extension does not have to be written by an attacker: Forever Security notes it could be malicious from the start, taken over through a bad update from its developer, or simply a legitimate extension with a vulnerability, which matters because many popular extensions request permission to read and change every site a user visits.
What the attacker needs
The attack needs three things: a browser extension allowed to run scripts on claude.ai, a user signed in to both claude.ai and Claude Desktop with Dispatch turned on, and a computer that is awake with the desktop app open. Dispatch only works while the computer is awake and the app is open, which limits the window but does not close it.
The reach goes further than one machine. Because Dispatch links a Claude account to every desktop app signed in to it, an extension running in the browser on one machine could target a different machine signed in to the same account. Forever Security says it has seen personal Claude accounts with cross-device Dispatch enabled on corporate endpoints, which is how the researchers frame the worst case: an extension on a personal laptop reaching a work laptop inside a company network.
The timeline: reported in May, still open in November
Forever Security reported the issue to Anthropic on May 14, 2026. Anthropic acknowledged the findings as valid on May 19, then closed the report as a duplicate of an issue it had already found internally. No CVE identifier was assigned and no bug bounty was paid, according to the researchers.
On October 13, Anthropic told the researchers it would deal with the problem by retiring Dispatch, targeting the end of October 2026, rather than shipping a patch. The Register reported on November 5 that Anthropic would not fix the flaw and did not answer its questions. The Claude help page, consulted in early November, still says Dispatch is in limited beta and that people already using it can continue, while being closed to new users. The researchers say the flaw is still exploitable.
The researchers add that the underlying design is not limited to Dispatch. They say Claude Code Remote Control and Claude Cowork sessions let extensions carry out the same attack, and they plan a follow-up post on the wider implications. That claim is theirs, not yet independently tested.
What to do now
First, check whether Dispatch is on. If you use Claude Desktop, open its settings and turn Dispatch off unless you actively use it to send tasks from your phone. Second, review which extensions can run scripts on claude.ai: in your browser's extension settings, look for any extension with permission to read and change data on all sites, and remove the ones you do not recognise or no longer need. Third, do not use a personal Claude account on a work computer with Dispatch enabled, since that is the cross-device path the researchers demonstrated. Finally, check Anthropic's current security notes before you decide anything, because the retirement status of Dispatch has already slipped once, from the end of October into November.
The Hacker News reported that the researchers had not observed real attacks of this kind. That is reassuring, but it is not a fix: the proof of concept shows the path exists, and the vendor's answer so far is to retire the feature rather than patch it.