Particle.news

Download on the App Store

Anthropic Pilots Claude for Chrome, Letting Its AI Click and Fill Forms in Your Browser

Anthropic is using a 1,000‑user research preview to validate defenses against browser‑specific prompt‑injection attacks before expanding access.

Image
Anthropic has cautiously entered the AI browser war.
Image
Image

Overview

  • Claude for Chrome launches as a controlled Chrome extension limited to 1,000 Max subscribers, with a waitlist for additional testers.
  • The agent runs in a side panel that keeps page context and can view websites, click buttons, and complete forms when granted permission.
  • Anthropic reports red‑team prompt‑injection success fell from 23.6% without mitigations to 11.2% with new safeguards, and from 35.7% to 0% on four browser‑specific attack types.
  • Built‑in protections include site‑level permissions, user confirmations for high‑risk actions, and default blocks on categories such as financial services, adult content, and pirated sites.
  • The preview positions Anthropic in a fast‑moving browser‑agent contest with Perplexity’s Comet, reported OpenAI browser efforts, and Google’s Gemini integrations; the Max plan is reported to cost between $100 and $200 per month.