A vulnerability in the Claude for Chrome extension lets malicious plugins control the AI

Security researchers have discovered a vulnerability in the Claude for Chrome browser extension developed by Anthropic. According to their findings, a malicious third-party extension can imitate user clicks and thereby trigger preset actions of the AI assistant without the user's knowledge.

What the problem is

Claude for Chrome gives the AI assistant access to the various web services a user is connected to, including Gmail, Google Docs, Google Calendar and Salesforce. The vulnerability that was found allows attackers to use that access against the user's intentions by simulating interaction with the extension's interface.

In this way, a malicious extension installed in the same browser can potentially force Claude to carry out actions that are normally initiated by the user deliberately — for example, ones related to processing mail, documents, calendar events or data in a CRM system.

Possible consequences

Because Claude for Chrome is integrated with work services and has access to confidential information, exploiting the vulnerability creates a risk of unauthorized access to a user's correspondence, documents and other data through the legitimate permissions granted to the AI assistant itself.

Vulnerabilities like this reflect a broader security problem with AI agents integrated into the browser: extensions with broad access rights to web services become an attractive target for attacks that exploit the particulars of how different extensions interact within a single browser.

Anthropic is the developer of Claude for Chrome, an extension that embeds the AI assistant's capabilities directly into the browser to automate the user's tasks.

Source: Bleeping Computer