Anthropic launches a browser-based Claude AI agent

Just a heads up, if you buy something through our links, we may get a small share of the sale. It’s one of the ways we keep the lights on here. Click here for more.

Anthropic announced a new research preview of Claude for Chrome, a browser-based AI assistant powered by its Claude models. 

For now, only about 1,000 people on Anthropic’s top-tier Max plan, costing $100 to $200 a month, will get access. But the company has also opened a waitlist for anyone else eager to try it.

The idea is straightforward: once installed as a Chrome extension, Claude sits in a side window, ready to answer questions, summarize pages, or even take actions in the browser, if the user allows it. 

Think of it as a co-pilot for your web browsing, capable of handling tasks so you don’t have to click through every form or menu yourself.

Anthropic’s move comes as AI labs increasingly see the browser as their next big battleground. 

Just a couple of months ago, Perplexity released Comet, a browser with an integrated AI agent, while Google has been weaving its Gemini assistant into Chrome. 

OpenAI is rumored to be working on a similar AI-powered browser, too. 

The stakes are high: with a federal judge expected to rule any day now on whether Google must spin off Chrome in its ongoing antitrust case, the future of who controls the most widely used browser suddenly feels wide open. 

Perplexity even offered $34.5 billion for Chrome, and Sam Altman of OpenAI hinted his company might make a bid as well.

Still, with great power comes new risks. Giving an AI direct access to your browser raises safety questions. 

Last week, Brave researchers flagged a flaw in Comet that could allow malicious sites to trick the AI into carrying out hidden commands, a so-called prompt injection attack. Perplexity says it’s been patched, but the incident highlights the stakes.

Anthropic says it’s building safeguards into Claude’s Chrome agent. By default, the AI won’t access sensitive categories like banking or adult sites. 

It also asks for explicit permission before high-risk actions such as publishing content or making purchases. The company claims its protections have already cut prompt injection success rates by more than half.

It’s not Anthropic’s first try at agentic AI. Last year, it tested a desktop-controlling agent that proved slow and clunky. 

But tools like Claude for Chrome suggest that AI helpers inside the browser are quickly becoming more capable, even if they’re still learning to handle the messier problems of the web.

Is giving AI agents direct browser access a natural evolution of digital assistants, or does it create too many security risks for everyday users? Should companies like Anthropic limit these powerful tools to premium subscribers, or make them widely available to accelerate AI adoption? Tell us below in the comments, or reach us via our Twitter or Facebook.