Skip to main content
Update

Claude in Chrome sessions now sync across devices, with a prompt injection warning attached

Claude in Chrome sessions now sync to your account instead of your device, carrying over to desktop and mobile, with a fresh warning about prompt injection risk.

By Nattapon YongpaiboonCo-founder, Claude Thailand Community

What’s changing

On 13 August 2026, Anthropic announced that the Claude in Chrome side panel now runs the same Claude Cowork session as the desktop, web, and mobile apps.

In practice, the session lives with your account, not with any single device or browser. Start a task in a Chrome tab, then pick it up on desktop, web, or mobile, and the conversation carries over in full. Skills and connectors you’ve set up also work inside the browser. Before this, browser sessions were completely isolated from everything else, and you had to start fresh every time you switched devices.

Illustration of a user account circle in the center, connected by lines to browser, desktop, and mobile icons, each marked with a checkmark, showing that one session now works continuously across all three because it’s tied to the account rather than the device

Who gets it

Available on Max and Team plans today, rolling out to Pro in the coming weeks.

The warning that came with it: prompt injection risk

On the same day, Anthropic separately warned that Claude in Chrome is a “browser agent,” which carries risk from prompt injection - instructions hidden in web content designed to trick Claude into taking actions the user never intended.

Per Anthropic’s “Use Claude in Chrome Safely” documentation, this is described as the single biggest risk facing browser-using AI tools. The current configuration reduces attack success rates to under 0.08% in Anthropic’s internal testing, but the article is explicit that the risk is not zero, and novel attacks may still emerge.

Two approval modes

  • Automatically approve - the side panel’s default. Claude screens its own actions and only pauses to ask when something needs approval.
  • Manually approve - you review every action before Claude executes it.

Anthropic recommends switching to Manually approve for higher-risk tasks or unfamiliar websites - it’s not a blanket recommendation to use that mode all the time.

Illustration of a mock webpage card with several lines of normal-looking content, one line highlighted with a dashed border and an orange warning badge overlapping it to represent a hidden instruction. A curved arrow points from the warning to a “Manual” mode pill, shown filled and checked, in contrast to a dimmer, unselected “Auto” pill beside it

Safety habits Anthropic recommends

  • Start with websites you trust; avoid unfamiliar ones.
  • Stay alert for suspicious behaviour, and stop the task immediately if Claude does something unexpected.
  • Use a separate browser profile without access to sensitive accounts.
  • Avoid using it for financial accounts, legal documents, medical information, or sensitive work data.
  • Don’t open the extension while viewing sensitive information on screen.

The writer’s take

The cross-device sync is genuinely convenient in my view, but once Claude becomes an agent that clicks around the web on its own, the risk is a different kind from a regular chat session. Personally, I’d keep Manually approve on, at least while you’re still getting a feel for it, especially on sites you don’t already trust.


Details in this article come from the Claude account on X (Twitter) and Anthropic’s “Use Claude in Chrome Safely” documentation, published 13 August 2026. Read the original post via the source link below.

Read the original >

Get it by email

New articles, Claude updates and community event announcements. Sent occasionally, never often enough to annoy you.

The newsletter is written in Thai

Carry on the conversation in our Facebook group

Ask questions, share techniques, show your work and hear about upcoming events. The group is where most of the talking happens.