Malicious Browser Extensions Can Exploit Claude Dispatch to Run Code on Users’ Computers
What Happened — Researchers at Forever Security demonstrated that a browser extension capable of sending scripts to claude.ai can trigger Claude Desktop’s Dispatch feature to execute arbitrary code on the victim’s machine. The flaw resides in the way Dispatch accepts tasks from external sources without sufficient validation of the caller’s origin. In a controlled test, the team loaded a malicious extension, issued a Dispatch request, and observed Claude Desktop launch a PowerShell payload that wrote a file to the user’s profile directory. Anthropic has acknowledged the issue and indicated that a fix is being prepared, but the vulnerability remains exploitable in current releases.
Why It Matters for Trust & Control Assurance — The incident tests the control objective of securing third‑party software components and enforcing strict extension vetting, a requirement that maps to multiple frameworks (e.g., NIST CSF Identify‑Protect).
Who Is Affected — SaaS AI providers with desktop integrations, enterprises that enable Claude Desktop, and any organization allowing users to install browser extensions that can interact with web‑based AI services.
Recommended Actions –
- Inventory and restrict browser extensions that can communicate with AI web portals.
- Enforce a policy that only approved extensions may access Claude Desktop or similar AI‑desktop bridges.
- Deploy monitoring for unexpected Dispatch calls and isolate Claude Desktop from untrusted browsers.
- Apply patches or updates from Anthropic as soon as they are released.
Source: The Hacker News
Technical Notes – Attack vector: malicious browser extension (malware) leveraging an unchecked Dispatch API to achieve local code execution (RCE). No CVE assigned yet; the flaw is a design‑level validation gap in Claude Desktop. Source: same as above