Anthropic is launching a research preview of a browser-based AI agent called Claude for Chrome, which is powered by the company’s Claude AI models. The agent will be available to 1,000 subscribers on Anthropic’s Max plan, priced between $100 and $200 per month. Users can access Claude through an extension in Chrome and interact with it in a sidecar window.
Anthropic aims to use this preview to identify and address new safety risks associated with AI agents accessing browsers. The company has introduced several defenses against prompt injection attacks, which have reduced the success rate of such attacks from 23.6% to 11.2%. Claude’s browser agent can limit access to certain sites, block adult content, pirated material, and financial services by default.
The research preview is part of a growing trend in AI labs developing browser integrations to offer seamless connections between AI systems and users. Other companies like Perplexity and OpenAI are also launching their own browsers with similar features. However, Anthropic warns that the rise of AI agents with browser access poses new safety risks, which need to be addressed.
Anthropic has previously launched an AI agent that could control your PC, but it was found to be slow and unreliable. The company’s latest AI model is more reliable, but many agentic systems still struggle with complex problems.
Source: https://techcrunch.com/2025/08/26/anthropic-launches-a-claude-ai-agent-that-lives-in-chrome