- Apr 30, 2026
AI lab Anthropic has launched a research preview of a new browser-based agent powered by its Claude AI model. On Tuesday, the company announced that “Claude for Chrome” will initially be available to 1,000 subscribers of their Max plan, which costs between $100 and $200 per month. A waiting list has also been opened for other interested users.
Selected users will be able to add an extension to Chrome and chat with Claude in a side window, which retains context from all activity happening in the browser. Users can allow Claude to perform certain tasks directly within the browser.
Browser integration has become a new competitive frontier for AI labs. Perplexity recently launched their own browser agent, Comet, capable of performing tasks on behalf of the user. OpenAI is reportedly planning a similar browser-based product soon, while Google has already integrated Gemini into Chrome.
Anthropic has warned that browser-based AI agents can create new security risks. Last week, Brave’s security team reported that Comet’s agent could be vulnerable to “indirect prompt injection” attacks. Anthropic says that their Claude agent has already implemented safety measures, reducing the success rate of such prompt injection attacks from 23.6% to 11.2%.
According to the company, Claude’s browser agent may restrict access to certain websites. It automatically blocks economic services, adult content, and counterfeit content. High-risk actions such as publishing, purchasing, or sharing personal information will require explicit user consent.
This new agent from Anthropic represents an improvement over their previous computer-controlled AI model. Tests around October 2024 showed the model to be slow and inconsistent, but Claude and other modern browser-based agents can now perform simple tasks reliably.
Amidst this AI-driven browser competition, excitement is mounting in the tech world due to Google’s potential Chrome-related initiatives and interest from Perplexity and OpenAI. Users can now complete daily tasks faster and more efficiently through AI agents.