2 min read
Anthropic has begun a research preview of Claude For Chrome that places an AI agent directly inside the Chrome browser. The launch opens access for a limited group of one thousand subscribers on Anthropic s Max plan and establishes a waitlist for other users. The extension creates a sidecar chat window that keeps context for the pages a user visits and can act in the browser when the user grants permission.
The agent can complete routine tasks such as drafting email text extracting information from open tabs and summarising pages. Users may allow Claude For Chrome to take low risk actions and must confirm before the agent performs high risk tasks such as publishing content making purchases or sharing personal data. Anthropic asks users to opt in for any task that could expose sensitive information.
The release comes as companies race to put agents into the browser. Competitors already ship browser agents or agent enabled browsers that automate tasks and handle multi step workflows. Anthropic notes that the browser as a platform creates new safety risks and it is using the preview to discover and fix new attack methods.
Anthropic has added several protective layers. The company blocks the agent by default from accessing sites that offer financial services adult content and pirated material. The agent will request explicit permission before acting on pages that pose higher risk. Anthropic also reports improved resistance to prompt injection attacks after engineering changes.
The company warned that agent capabilities are not perfect and that complex tasks may still fail. The research preview is meant to gather real world feedback and to harden safety controls. Anthropic will study how Claude For Chrome performs on daily tasks and how its safeguards hold up in diverse browsing contexts.
