Anthropic is testing its AI-powered browser extension, Claude for Chrome, with 1,000 subscribers who pay $100 or $200 per month. The extension allows the company’s machine learning model to automate web browsing, but it also poses significant security and privacy risks.
Browser extensions have access to sensitive information and often request broad permissions, making them vulnerable to abuse. Anthropic’s Claude is no exception, with a warning about unintended actions, probabilistic behavior, and potential data breaches.
The extension does implement a permission system, but it also offers a high-risk mode for fully autonomous operation. This raises concerns about computer security when generative AI models are added to the mix.
Despite these risks, Anthropic argues that browser-using AI is inevitable and will make web browsing more useful. The company claims to have reduced the success rate of prompt injection attacks by 12 percentage points after adding safety mitigations. However, it won’t release Claude for Chrome to the general public until security improves.
Source: https://www.theregister.com/2025/08/26/anthropic_claude_chrome_warnings