Anthropic launches a Claude AI agent that lives in Chrome

Anthropic is launching a analysis preview of a browser-based AI agent powered by its Claude AI fashions, the corporate announced on Tuesday. The agent, Claude for Chrome, is rolling out to a bunch of 1,000 subscribers on Anthropic’s Max plan, which prices between $100 and $200 per thirty days. The corporate can be opening a waitlist for different customers.

By including an extension to Chrome, choose customers can now chat with Claude in a sidecar window that maintains context of every part occurring of their browser. Customers may give the Claude agent permission to take actions of their browser and full some duties on their behalf.

The browser is rapidly turning into the following battleground for AI labs, which intention to make use of browser integrations to supply extra seamless connections between AI techniques and their customers. Perplexity lately launched its personal browser, Comet, which options an AI agent that may offload duties for customers. OpenAI is reportedly near launching its personal AI-powered browser, which is rumored to have related options to Comet. In the meantime, Google has launched Gemini integrations with Chrome in current months.

The race to develop AI-powered browsers is particularly urgent given Google’s looming antitrust case, during which a remaining choice is anticipated any day now. The federal decide within the case has advised he could pressure Google to promote its Chrome browser. Perplexity submitted an unsolicited $34.5 billion offer for Chrome, and OpenAI CEO Sam Altman advised his firm can be willing to buy it as properly.

Within the Tuesday weblog submit, Anthropic warned that the rise of AI brokers with browser entry poses new security dangers. Final week, Courageous’s safety workforce mentioned it discovered that Comet’s browser agent might be susceptible to indirect prompt-injection attacks, the place hidden code on an internet site may trick the agent into executing malicious directions when it processed the web page.

(Perplexity’s head of communications, Jesse Dwyer, instructed TechCrunch in an e-mail that the vulnerability Courageous raised has been fastened.)

Anthropic says it hopes to make use of this analysis preview as an opportunity to catch and handle novel security dangers; nevertheless, the corporate has already launched a number of defenses in opposition to immediate injection assaults. The corporate says its interventions decreased the success price of immediate injection assaults from 23.6% to 11.2%.

For instance, Anthropic says customers can restrict Claude’s browser agent from accessing sure websites within the app’s settings, and the corporate has, by default, blocked Claude from accessing web sites that provide monetary providers, grownup content material, and pirated content material. The corporate additionally says that Claude’s browser agent will ask for consumer permission earlier than “taking high-risk actions like publishing, buying, or sharing private knowledge.”

This isn’t Anthropic’s first foray into AI fashions that may management your laptop display screen. In October 2024, the corporate launched an AI agent that would control your PC — nevertheless, testing on the time revealed that the mannequin was fairly sluggish and unreliable.

The capabilities of agentic AI fashions have improved fairly a bit since then. TechCrunch has discovered that trendy browser-using AI brokers, akin to Comet and ChatGPT Agent, are pretty dependable at offloading easy duties for customers. Nonetheless, many of those agentic techniques nonetheless wrestle with extra advanced issues.