Claude in Chrome: Your Session Escapes the Tab

Start beside a signed-in page, then pick up the same conversation elsewhere without rebuilding its context. Claude says conversations are saved and its skills and connectors can work in the browser.

That continuity makes browser work feel less disposable. It also asks more of our judgement, because the session can see and act beside the pages we choose to open.

Claude says it can read what is visible on a signed-in page, click, type and fill forms. That can make a real web task easier to continue, but it changes the question from “can this chat help?” to “what should this session be allowed to reach?”

Anthropic says prompt injection is the biggest risk for browser-using AI tools and that the risk is not zero. Convenience is useful only when human control remains part of the work.

Claude’s launch post names desktop, web and mobile as the places a Chrome session can continue. A related Claude post puts the point more plainly: the side panel runs the same Cowork session, and that session lives with the account rather than a single device.

The details matter. The launch post described Max and Team access with Pro rolling out, while product and Help Centre wording describes a broader paid-plan position with further conditions. Availability is not a timeless claim; it needs checking when this is published.

The account-level design is the practical change. We can begin a task in a tab, retain the conversation and return to it from another Claude surface rather than treating each device as a fresh start.

That does not make every page suitable for browser action. It means the context we preserve, the permissions we grant and the pages we expose travel together as part of one working session.

Sessions live with your account, not on any single device, so you can start in a tab and pick it up later somewhere else.

Claude

The visual evidence is useful because it is specific: Claude depicts its browser tool operating beside a calendar workflow. Alongside its product and Help Centre material, it supports the narrower claim that Claude is designed to read visible pages and perform approved browser actions.

It cannot settle the harder questions for us. Whether an account has access, whether an organisation has enabled it, and whether a page contains information we should keep out of reach remain live decisions.

Claude describes a browser side panel that works beside the page: it can read what is visible, click, type and fill forms. The session can then be saved in account history and continued across Claude surfaces.

The important word is control. We still decide which page to open, what to expose and when an action should proceed. Continuity should preserve context, not bypass judgment.

A simple task begins with less friction: we open Claude beside the page, establish the context and keep the conversation when we leave the tab. Later, that same session can be resumed on desktop, web or mobile rather than reconstructed from fragments.

The experience will not look identical for every account. Claude’s rollout and plan wording is time-sensitive, enterprise access can require enablement, and browser action deserves extra care on pages carrying sensitive information.

A session that follows the account is more useful than a disposable browser chat—and more deserving of deliberate control.

Creator Broadcast synthesis from Claude’s product and safety guidance.

Claude’s account-level continuity is a sensible answer to a familiar frustration: work begun in one place should not have to be rebuilt somewhere else. The value is real when a browser task needs the context already established on another surface.

But continuity is not a free pass for browser action. Anthropic’s own safety guidance keeps the boundary clear: prompt injection remains a risk, and visible sensitive information can be exposed to the session. The stronger habit is to use the continuity while keeping each permission decision explicit.