Anthropic Launches Browser-Based AI Agent Powered by Claude AI Models
Anthropic has announced the launch of a research preview for a browser-based AI agent powered by its Claude AI models. The new agent, Claude for Chrome, is being rolled out to a select group of 1000 subscribers on Anthropic’s Max plan, which ranges in price from $100 to $200 per month. Interested users can also join the waitlist for access to Claude for Chrome.
New Features of Claude for Chrome
By adding an extension to Google Chrome, users can now interact with Claude in a sidecar window that keeps track of all browser activity. Users can grant the agent permission to perform tasks on their behalf and take actions within the browser.
Competition in the AI-Powered Browser Space
The integration of AI into web browsers is becoming increasingly popular among AI labs. Perplexity recently introduced Comet, a browser with an AI agent capable of assisting users with various tasks. OpenAI is also rumored to be close to launching its own AI-powered browser with features similar to Comet. Google has already integrated Gemini features into Chrome in recent months.
AI Browsers Amid Google’s Antitrust Concerns
The development of AI-powered browsers is gaining momentum, especially in light of Google’s impending antitrust case. With a potential decision on the horizon, there is speculation that Google may be required to sell its Chrome browser. Perplexity has made a bid for Chrome, while OpenAI’s CEO has expressed interest in acquiring it as well.
Safety Concerns Surrounding AI Agents in Browsers
Anthropic has raised concerns about the security risks associated with AI agents accessing web browsers. Recent findings by Brave’s security team revealed potential vulnerabilities in Comet’s browser agent. Anthropic has taken measures to address these risks, reducing the success rate of prompt injection attacks from 23.6% to 11.2%.
Enhanced Safety Measures for Users
Anthropic has implemented several safeguards to protect users while using Claude for Chrome. Users can restrict the agent’s access to certain websites in the app’s settings. Claude is also blocked from accessing sites offering financial services, adult content, and pirated material by default. Additionally, the browser agent seeks user permission before executing high-risk actions like sharing personal data.
Continued Innovation in AI Models
Anthropic’s latest venture into AI models controlling web browsers follows its previous release of an AI agent for PC control in 2024. While initial testing revealed performance issues, advancements in agentic AI models have significantly improved reliability. Modern browser-using AI agents like Comet and ChatGPT Agent are now adept at handling simple tasks, although challenges remain with more complex operations.