Security researchers at Manifold Security discovered two vulnerabilities in Anthropic's Claude for Chrome extension that remain unpatched months after being reported. The flaws allow malicious extensions to exploit AI privileges. The researchers found the issues in the extension.
The vulnerabilities enable rogue extensions to perform privileged AI actions, including accessing Gmail and Google Docs. This is possible because the Claude extension fails to properly validate requests from other extensions. As a result, attackers can manipulate the extension into taking unauthorized actions.
The researchers at Manifold Security noted that the Claude extension's architecture allows it to interact with other Google services, making it a valuable target for attackers. By exploiting the vulnerabilities, malicious extensions can gain access to sensitive user data.
The flaws were reported to Anthropic, but the company has not yet released a patch. The researchers are concerned that the vulnerabilities could be exploited by malicious actors, potentially leading to data breaches or other security incidents.
The discovery of these vulnerabilities raises questions about the security of AI-powered browser extensions. As AI becomes increasingly integrated into browser functionality, the potential attack surface grows.
The consequences of these vulnerabilities are significant, as they could allow attackers to gain access to sensitive user data. Users of the Claude extension should be cautious when installing other extensions, as malicious software could exploit these flaws.
What are the risks associated with these vulnerabilities? The risks include unauthorized access to Gmail, Google Docs, and other sensitive user data. Users may experience data breaches or other security incidents. How were the vulnerabilities discovered? Manifold Security researchers found the flaws while investigating the Claude extension's architecture and functionality. What can users do to protect themselves? Users should be cautious when installing new extensions and monitor their browser activity for suspicious behavior.