Anthropic Launches Claude AI Agent for Chrome

Anthropic has launched a research preview of its Claude AI agent for Chrome, allowing users to interact with the AI directly within their browser. This new feature is initially available to 1,000 subscribers on Anthropic's Max plan, with a waitlist open for additional users. The Claude for Chrome extension enables users to chat with the AI in a sidecar window, maintaining context of their browser activities and even allowing the AI to perform certain tasks on their behalf announced on their website.

The introduction of browser-based AI agents is becoming a competitive area for AI labs, with companies like Perplexity and OpenAI also developing similar technologies. Anthropic's Claude for Chrome aims to enhance user productivity by integrating AI capabilities directly into the browser, though it acknowledges the potential safety risks involved. The company has implemented several defenses against prompt injection attacks, reducing their success rate from 23.6% to 11.2%.

Anthropic is using this research preview to gather real-world feedback and improve the safety and functionality of Claude for Chrome. Users can control the AI's access to specific websites and must grant permission for high-risk actions. The company plans to expand access as it strengthens safety measures and gains more insights from this limited rollout.