Skip to content- Sources: primary
- Summary: Forever Security researcher Gal Weizman showed that a content script on claude.ai can find the Claude Desktop Dispatch session through the API with the user's cookies, send prompts, and auto-click Allow until the agent runs code. Anthropic accepted the report in May as a duplicate, paid no bounty, issued no CVE, and said it would retire Dispatch at the end of October, which has not happened. Weizman says the same weakness applies to Claude Code Remote Control and Cowork sessions.
- Why it matters: Organizations should restrict browser extensions on claude.ai and keep personal Claude accounts off work machines running Claude Desktop.
- Follow-up: Anthropic retiring Dispatch or changing cross-device session trust for Cowork and Remote Control.