Distinguished privacy expert Alexander Hanff recently published a profound analytical discourse on his blog, unveiling that Claude Desktop surreptitiously deploys a native messaging bridge component during its installation. This bridge facilitates communication between specific browser extensions and local executables, enabling elevated privileges such as browser automation, DOM accessibility, and session sharing—all executed without the explicit consent of the user. (Note: These findings pertain exclusively to the desktop client and bear no relation to Claude Code.)
Hanff discovered a JSON manifest within his Brave browser that he had not personally installed. This file designated the path for a localhost executable and preemptively authorized three specific extension IDs to invoke the host and perform operations. He confirmed that no Claude browser extensions were present on his system, yet this manifest was generated autonomously, permitting extensions to operate outside the browser’s sandbox at the user’s privilege level. This essentially inaugurates a communication conduit between the browser and local applications.
According to official documentation, once this bridge is activated, it can facilitate browser automation (sharing login states and accessing any authenticated site), real-time monitoring of console errors and DOM states, structured data extraction, form population, and the recording of browser sessions as GIF animations.
Such capabilities imply that extensions could potentially access sensitive sessions involving banking, healthcare, or corporate systems, while simultaneously reading decrypted page content. Hanff’s audit revealed that this pre-authorized bridge remains in a state of readiness even if the extensions are not actively utilized. Consequently, should a prompt injection attack occur, an adversary could exploit this component to surveil users and exfiltrate sensitive data, posing a formidable threat to digital privacy and security.
Hanff characterizes this behavior as the clandestine installation of a browser surveillance component, citing several egregious practices:
- Forced Bundling: Modification of browser configurations across seven Chromium-based browsers without user authorization.
- Absence of Opt-out: The installation and execution proceed without any notification, depriving users of the ability to decline.
- Persistent Presence: The manifest file is automatically rewritten upon each launch, rendering it impossible to remove unless the Claude desktop client is entirely uninstalled.
- Obscure Pre-authorization: The identities and functions of the pre-authorized extensions listed in the manifest remain enigmatic.
- Deceptive Nomenclature: The manifest is misleadingly named after a Claude browser extension to obfuscate its true nature.
- Excessive Directory Creation: Relevant directories and manifest files are generated even if the specific browsers are not installed on the system.
- Understated Scope: While documentation claims support only for Chrome and Edge, the component actually affects seven distinct Chromium browsers.
- Lack of Interface Transparency: No user interface exists to inform the user that the local machine has been registered as an executable host.
To date, no official rejoinder has been issued. While the component’s functionality is ostensibly designed to allow Claude to orchestrate the browser, the quintessential issue remains the lack of user mandate and the opaque nature of its deployment. This bridge potentially subverts established browser security mechanisms, creating a latent risk that cannot be mitigated unless the Claude desktop application is removed from the system.
Support Our Threat Intelligence
If you find our CVE report and cybersecurity news helpful, consider supporting our work.