Claude AI Agent Now Available in Chrome Browser

Anthropic announced on Tuesday a research preview of a browser-based AI agent powered by its Claude models. Named Claude for Chrome, the agent is being made available to 1,000 subscribers on Anthropic's premium Max plan, priced from $100 to $200 monthly. The company has also opened a waitlist for other interested users.

By installing a Chrome extension, selected users can interact with Claude in a dedicated sidebar that maintains awareness of their browser activity. Users can also authorize the Claude agent to perform actions and complete certain tasks on their behalf directly within the browser.

The browser is rapidly emerging as the next major arena for AI labs, which are leveraging browser integrations to create more seamless connections between AI and users. Perplexity recently launched its Comet browser featuring an AI agent capable of handling user tasks. OpenAI is also reportedly near launching its own AI-powered browser with features similar to Comet. Simultaneously, Google has been integrating its Gemini AI with Chrome in recent months.

The push to develop AI-powered browsers is intensifying, partly due to Google's impending antitrust case, where a final ruling is expected imminently. The presiding judge has indicated he might compel Google to divest its Chrome browser. Perplexity has made an unsolicited $34.5 billion offer for Chrome, and OpenAI's CEO Sam Altman has suggested his company would also be interested in a purchase.

In its Tuesday announcement, Anthropic cautioned that AI agents with browser access introduce new safety concerns. Last week, Brave's security team reported finding a vulnerability in Comet's browser agent to indirect prompt-injection attacks, where hidden website code could manipulate the agent into executing malicious instructions.

(Jesse Dwyer, Perplexity's head of communications, informed TechCrunch via email that the vulnerability identified by Brave has been remediated.)

Anthropic states that this research preview aims to identify and mitigate novel safety risks. The company has already implemented several defenses against prompt injection attacks, claiming its measures have reduced the success rate of such attacks from 23.6% to 11.2%.

Tech and VC Leaders Join the Disrupt 2025 Lineup

Netflix, ElevenLabs, Wayve, Sequoia Capital, and Elad Gil are among the prominent figures featured on the Disrupt 2025 agenda. They will share key insights to drive startup growth and refine your competitive strategy. Join the 20th anniversary of TechCrunch Disrupt for an opportunity to learn from leading tech voices — secure your ticket now and save over $600 before prices increase.

Tech and VC Leaders Join the Disrupt 2025 Lineup

Netflix, ElevenLabs, Wayve, and Sequoia Capital are just some of the influential names joining the Disrupt 2025 agenda. They are set to deliver the crucial insights that propel startup growth and enhance your strategic advantage. Don't miss the 20th anniversary of TechCrunch Disrupt and this chance to learn from industry leaders — get your ticket today and save up to $675 before rates go up.

For instance, Anthropic notes that users can restrict Claude's browser agent from accessing specific sites within the app settings. By default, the company has blocked access to websites offering financial services, adult content, and pirated material. Anthropic also states that Claude's browser agent will request user consent before undertaking high-risk actions such as publishing content, making purchases, or sharing personal data.

This is not Anthropic's first venture into AI models capable of interacting with your computer interface. In October 2024, the company released an AI agent designed to control a user's PC, though initial testing revealed the model was relatively slow and inconsistent.

The capabilities of agentic AI models have advanced significantly since then. TechCrunch has observed that contemporary browser-based AI agents like Comet and ChatGPT Agent are reasonably reliable at handling simple user tasks. However, many of these agentic systems continue to face challenges with more complex multi-step problems.