Anthropic launches Claude for Chrome, a browser-based AI agent now in research preview for select users.
It's available to 1,000 subscribers on its premium Max plan, costing $100-$200/month, with a waitlist for others.
The Chrome extension opens a sidecar window that maintains context of your browser activity.
Users can grant it permission to take actions and complete tasks within the browser on their behalf.
This launch is part of a new battleground for AI labs, competing with Perplexity's Comet and OpenAI's rumored browser.
The race intensifies with Google's looming antitrust case, which may force a sale of Chrome.
Anthropic warns such agents pose new safety risks, like indirect prompt-injection attacks.
It has introduced defenses, reducing prompt injection success rates from 23.6% to 11.2%.
By default, Claude is blocked from accessing financial, adult, and pirated content sites.
While modern AI agents are reliable for simple tasks, they still struggle with complex problems.