Anthropic has launched a research preview of a browser-based AI agent powered by its Claude models, the company announced Tuesday. The tool, called Claude for Chrome, is rolling out to 1,000 subscribers on Anthropic’s Max plan, which costs between $100 and $200 per month, with a waitlist open for others.
The extension allows users to chat with Claude in a sidebar that maintains context of their browsing activity. With permission, Claude can also take actions in the browser, enabling it to complete tasks on behalf of users.
The move highlights the browser as a key battleground for AI labs. Perplexity recently launched its own AI-powered browser, Comet, while OpenAI is reportedly preparing a similar product. Google has also integrated its Gemini models into Chrome.
Anthropic acknowledged that browser-based AI agents introduce new safety risks. Last week, Brave warned that Comet’s agent was vulnerable to prompt-injection attacks, though Perplexity has since patched the issue. Anthropic said it has already deployed safeguards, reducing the success rate of such attacks from 23.6% to 11.2%. By default, Claude is blocked from accessing financial services, adult sites, and pirated content, and it will always ask before taking high-risk actions such as purchases or sharing personal data.
Related: Anthropic Gives Claude AI Power to End Conversations in Extreme Cases
This is not Anthropic’s first agent. In October 2024, the company tested an AI system that could control PC screens, though it was limited by speed and reliability. Since then, agentic models have advanced significantly, with systems like Claude, Comet, and ChatGPT Agent increasingly capable of handling everyday tasks, though many still falter with complex workflows.
