Anthropic has launched a research preview of its new Claude for Chrome agent, giving select subscribers access to an AI assistant that lives inside the world’s most popular browser.

The extension is rolling out to 1,000 Max plan customers (priced between $100 and $200 per month), with a waitlist now open for other users.

What Claude for Chrome can do

With the new Chrome extension, users can:

Chat with Claude in a persistent sidecar window that maintains context of their browsing session.
Authorize Claude to take actions directly in the browser, such as filling out forms or completing tasks.

The move positions Anthropic alongside rivals racing to bring AI deeper into daily internet use.

Browser becomes AI’s next battleground

Anthropic isn’t alone in eyeing the browser as prime real estate for AI agents:

Perplexity recently launched its own browser, Comet, which comes with a built-in AI assistant.
OpenAI is rumored to be preparing its own AI-powered browser with features similar to Comet.
Google has rolled out early Gemini integrations with Chrome, even as the company faces mounting antitrust pressure.

The competition is intensifying as a U.S. federal judge considers whether Google may be forced to sell Chrome. Perplexity has already floated a $34.5 billion bid for the browser, while OpenAI CEO Sam Altman has publicly hinted at buying it too.

Safety concerns with browser AI

Alongside the rollout, Anthropic highlighted the new security risks AI browser agents introduce.

Prompt injection attacks: Malicious websites can embed hidden instructions that trick an AI into executing unintended actions.
Case in point: Last week, Brave’s security team flagged such vulnerabilities in Perplexity’s Comet. (Perplexity says the issue has since been fixed.)

Anthropic claims it has implemented safeguards that reduced the success rate of these attacks from 23.6% to 11.2% in testing.

Anthropic’s experiment underscores the growing push to make AI agents “live” inside browsers, where they can directly handle online tasks for users. But as the technology spreads, so do the risks — forcing AI companies to balance convenience with security.

For now, Claude for Chrome is limited to a small test group, but if successful, it could become one of the most widely used AI agents in daily digital life.