Meta puts its Muse agent in a Chrome side panel: 200,000 users already
Meta has released a Chrome extension that opens its Muse AI agent in the browser's side panel. The agent navigates websites and completes tasks on its own, but it works only within its own tab group and cannot see saved passwords, browsing history, or Chrome settings. Alexandr Wang reported that 200,000 people already use the extension. For developers and product teams, it's another sign that the race to build AI agents has moved directly into the browser.
AI-processed from TNW; edited by Hamidun News
Browser agents long looked like conference demos: impressive, but a bit frightening to hand your Chrome, with all its logins, over to. Meta has taken a step that shows how companies are trying to ease that fear. Meta's chief AI officer, Alexandr Wang, announced on X a Muse extension for Chrome that opens the agent in the browser's side panel. According to him, 200,000 people already use the extension.
What Muse can do in the browser
Muse runs inside the user's regular Chrome. You open the side panel, give the agent a task, and it navigates pages on its own, reads them, and performs actions on websites. It's like an assistant sitting next to you at your laptop, clicking the mouse by itself while you get on with other things.
An important detail: the agent acts only inside its own tab group. It doesn't poke into your other pages, which means it can't accidentally read your work email open in another tab unless you explicitly ask it to.
What boundaries Meta has set
According to the description on the Chrome Web Store, Muse cannot see saved passwords, browsing history, or Chrome settings. For an extension, these are baseline guarantees: even if the agent makes a mistake or is misled by a malicious page, it won't have access to the browser's most sensitive data.
At the same time, Meta does not hide the limitations. In the same extension listing, the company warns that Muse may take unexpected actions. There is another caveat: the agent's activity may appear in the user's Google account. To websites and services, such actions look like an ordinary human session, and it may be hard to later tell what the user did and what the agent did.
This is an honest admission: agents built on large language models still cannot be trusted unconditionally. Restricting the agent to tabs and keeping passwords out of reach lower the risk, but do not eliminate it entirely.
Why this matters for developers and product teams
First, 200,000 users in a short time confirm demand for agents that work where people already spend their day, which is the browser. There's no need to install a separate app or learn a new interface; an extension and a side panel are enough.
Second, Meta chose a familiar isolation model: the agent gets a sandbox made of its own tabs rather than access to the entire profile. If you're building your own browser agent, this is a useful reference point. A separate workspace for tasks, locked-away secrets, and a warning about possible errors look like a reasonable minimum.
Third, shipping a Chrome extension means Meta is competing with Google on Google's own turf. One company's agent lives inside another company's browser, and that browser sets the rules: what permissions to grant, what to show in the account, how to tell human actions from program actions. For small teams, this is a reminder that platform dependence remains a risk, even when a product is growing fast.
What remains unclear
Meta has not disclosed which tasks users most often hand to Muse or how often the agent makes mistakes. The figure of 200,000 users comes from Wang's statement and has not been independently verified. It is also unknown whether it counts active users or just installs.
The practical takeaway is simple. If you're experimenting with browser agents, give them tasks with a low cost of error: searching, gathering information, comparing options. Purchases, sending emails, and changing settings are better left under manual confirmation until the industry can reliably explain what an agent did on your behalf.
- Nov 21, 2026Meta's Muse agent mixes up other people's chats and shows employees bank balances
- Nov 21, 2026llama.cpp Hits 1.0.0: Breaking the C API Now Requires a Major Version Bump
- Nov 21, 2026Hugging Face CEO Vows to Defend Open AI After Huang Exits Nvidia
- Nov 21, 2026A coding model that beat April's leader got 20x cheaper by October
Key Questions Answered
What Muse can do in the browser?
Muse runs inside the user's regular Chrome. You open the side panel, give the agent a task, and it navigates pages on its own, reads them, and performs actions on websites. It's like an assistant sitting next to you at your laptop, clicking the mouse by itself while you get on with other things. An important detail: the agent acts only inside its own tab group. It doesn't poke into your other pages, which means it can't accidentally read your work email open in another tab unless you explicitly ask it to.
What boundaries Meta has set?
According to the description on the Chrome Web Store, Muse cannot see saved passwords, browsing history, or Chrome settings. For an extension, these are baseline guarantees: even if the agent makes a mistake or is misled by a malicious page, it won't have access to the browser's most sensitive data. At the same time, Meta does not hide the limitations. In the same extension listing, the company warns that Muse may take unexpected actions.
Why this matters for developers and product teams?
First, 200,000 users in a short time confirm demand for agents that work where people already spend their day, which is the browser. There's no need to install a separate app or learn a new interface; an extension and a side panel are enough. Second, Meta chose a familiar isolation model: the agent gets a sandbox made of its own tabs rather than access to the entire profile. If you're building your own browser agent, this is a useful reference point.
Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews
Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.