Muse Moved Onto Your Face. The Beta Gives You Hands-Free Help and Takes Away Your Old Assistant.

Picture this. You are standing in a grocery aisle, both hands full, staring at two jars of pasta sauce, and you say a name out loud. Not "Hey Meta." A name you picked. Your agent looks through the camera on your glasses, reads the labels, checks your calendar for tonight's dinner guests, and tells you which jar to buy. No phone, no unlocking, no typing with your thumbs while a jar slides off your arm.

That is the pitch, and as of November 6, 2026, it is a real beta you can turn on. Muse, Meta's personal AI agent, now runs on Meta's AI glasses. Meta announced it at Connect in September as "coming months," then shipped it about six weeks later, which by Meta's standards counts as fast. More than 10 million people in the US and Canada had already set up Muse agents by the time of Meta's October 21 earnings call, and about 25 percent of them use it on any given day, so the glasses beta lands on an agent with a daily audience of roughly 2.5 million people. That last number is my arithmetic on Zuckerberg's rounded figures, not a number Meta published, but the scale is the point. This is not a lab demo. It is a product with millions of daily users getting a new body.

Here is the catch, and it is a weird one. Switching your glasses to Muse is one toggle in the Meta AI app. Glasses, device settings, AI assistant, pick Muse. Thirty seconds. But the toggle does not add Muse alongside Meta AI; it replaces it. Only one assistant is active on your glasses at a time, and the two of them do not share anything. Your connected services do not carry over, and neither do your permissions. Each app asks for its own, separately, which means the thirty-second switch comes with a second, unadvertised job: reconnecting everything you already connected, one service at a time, inside a different app.

What you actually gain

Start with the good stuff, because there is real good stuff here. Meta AI on glasses answers questions while Muse runs errands. That is the difference that matters, and it is the whole reason the beta exists. You hand off a task with your voice and the agent keeps working after you stop talking, the same way it does in the Muse app on your phone, except now you never took the phone out. At Connect, Meta demoed guided workouts, meal logging, appointment booking, and shopping help for items sitting right in front of the wearer. Look at a shelf, ask about it, get an answer that knows what you are looking at.

You also get to name it. Muse is personalized, so you can give your agent a name and record it, then summon it by that name. "Hey Muse" works too, and so does "Hey Meta," though custom names and "Hey Muse" are not available on every glasses model during the beta, and Meta does not publish a model-by-model list. Beta participants get extra usage during the beta period, on top of the free tier with its usage limit. Subscriptions exist for more: Power at $20 a month, Maximum at $100.

None of that is hypothetical. It is in Meta's own help documentation, which I read in full. What Meta does not do anywhere is tell you the price of the switch in the same place it tells you the benefits. So I counted it.

The ledger

Read the middle rows again. Music, calling, messaging, navigation: the four things people actually do with glasses all day are the four things the beta does worst. Meta's own wording is vague about the extent, saying only that "a few features, including some" of each are not supported yet, and it has not published which ones, a vagueness that is itself a finding. You cannot plan around a gap you cannot measure. If you need navigation on a drive, switch back to Meta AI before you leave, because discovering the gap at a highway interchange is a bad way to learn what "beta" means.

And the switch back is the same toggle, which cuts both ways. Reversibility is the beta's best feature. Nothing here is permanent. But every flip means living under a different privacy policy for your voice, because the policy follows the assistant, not the hardware. Same microphone, same glasses, two rulebooks, and the wearer is expected to keep track of which one is in force.

A microphone with no light

Now the part Meta does not put in the help page. Ray-Ban Meta Audio, the $349 camera-free pair Meta launched at Connect, has a six-microphone array and no indicator light for audio recording. Camera glasses carry an LED that tells bystanders when they are filming. Microphones have no equivalent signal, and since Muse on glasses waits for a wake word, the device has to listen continuously to ambient audio. People nearby cannot tell waiting from recording. IDC analyst Jitesh Ubrani put it plainly in coverage of the launch: the problem is not that the glasses record, but that nobody around you can tell when they do.

That distinction stopped being theoretical on October 13, 2026, when a Ray-Ban Meta owner reported on Reddit that three different pairs of glasses, second and third generation, had saved audio recordings and text transcripts of ordinary conversations to the Meta AI app without anyone saying the wake phrase, even with the "Respond without Hey Meta" setting turned off. Meta told Android Authority the same day that the issue had been resolved and that people could review and delete voice logs in settings. Meta named no cause, no firmware version, and no count of affected devices. Android Authority said it had not reproduced the behavior. A later reply from the original owner said a further update had not stopped it on their glasses, which is disputed and unverified, not proof of anything except that the story is not closed.

I am not saying your glasses are recording you right now. I am saying the company that sells a microphone with no light also sells you a wake-word system that misfires, and it told you so itself. Meta's help page warns that a voice interaction can be misactivated even if you do not say your custom name, and that false wakes may be more frequent if your custom name sounds like another commonly used word. Pick a name like "Sam" or "Alex" and you have built yourself a device that wakes up when strangers say hello to each other.

The false-wake clock

Here is a piece of arithmetic nobody at Meta has volunteered. When Meta's systems detect that an activation was not intended, they label it a false wake and delete it within 90 days of detection. Other voice transcripts and audio are kept for up to one year. So a misactivation that Meta catches gets deleted in three months, while a misactivation Meta fails to catch can sit in storage for a full year, four times longer, and you have no way to know which bucket any given recording fell into. Your only audit tool is the voice activity log in the Meta AI app, which you have to open and check yourself.

It gets stranger across the Atlantic. Meta publishes its AI Glasses Voice Privacy Notice in regional versions, and the versions do not describe the same product. In the United States, the notice says text transcripts and audio recordings of your voice interactions are stored by default to help improve Meta's products, and it offers no setting that turns storage off. In the United Kingdom, the same notice says you can choose to store your voice recordings. Same glasses, same microphone, opposite defaults, separated by an ocean. The US notice also says Meta will store voice recordings even if you unintentionally activate a voice interaction. That sentence is doing a lot of work, and it is worth reading twice.

The other side, at full strength

In fairness, this beta is the right shape for a personal agent, and the glasses are the right body for it. Every other way of talking to Muse starts with friction: unlock the phone, open the app, describe the situation in words. Glasses remove all three steps at once, and that removal is the difference between an agent you use for big planned tasks and an agent you use for the small ones that actually fill a day. Meal logging, a quick question about the thing in your hand, a task handed off while your hands are full: these are the jobs where a phone in your pocket is not a minor inconvenience but the reason you never delegate at all. And the gaps are the features Meta AI already does well, one toggle away, which makes the beta less of a commitment than it looks. Thirty seconds to switch in, thirty seconds to switch back, extra usage thrown in for beta testers. Treating unfinished beta features as permanent defects would be scoring a rehearsal like opening night.

What I could not verify

Honesty requires the gaps. I have not tested the beta on hardware, so every capability claim here comes from Meta's documentation and Connect demos, not from wearing the glasses myself. Meta has not published which music, calling, messaging, or navigation features are specifically missing, so the ledger above is as precise as the public record allows, which is not very precise. The October 13 recording reports come from one owner and community threads; Meta says the issue is resolved, the owner's later claim that it persisted is unverified, and nobody has published a cause, a firmware version, or an affected-device count. The US and UK voice-storage comparison comes from a side-by-side reading of the regional notice pages, not from any statement by Meta acknowledging the difference. And no independent data exists yet on how often Muse on glasses misactivates compared with Meta AI, so the false-wake warning is Meta's own language, not a measured rate.

What to do before you flip the toggle

First, list the services you actually use through Meta AI on your glasses, then reconnect each one in the Muse app before you need it, because nothing carries over and you will discover the missing connection at the worst moment. Second, pick a custom name that sounds like nothing else you say in a normal day, not a common first name and not a word that shows up in small talk, or your false wakes will multiply exactly the way Meta warns they will. Third, open the voice activity log in the Meta AI app after your first week and delete anything you did not intend, because the log is the only audit trail you get and nobody will check it for you. Fourth, if you need navigation or calls, switch back to Meta AI before you leave the house, not after you are already lost. Fifth, keep the stakes low while the beta is young: errands, lookups, meal logs, the kind of task where a misheard instruction costs you a wrong jar of sauce and not a wrong flight. And if you wear the camera-free Audio pair, remember the people around you cannot see your microphone working. Say the name out loud before you start talking to your agent, so the humans in the room know who you are talking to. That courtesy costs nothing, and the glasses cannot provide it for you.

Muse on your face is a genuine step toward the agent Meta has been describing since September: a helper you talk to instead of a screen you operate. But a beta that swaps your assistant, drops four daily features, re-permissions your services, and changes your privacy policy with one toggle is not a free upgrade. It is a trade, and the honest version of the trade is printed above, in a table Meta never published. Read it before you flip the switch, because the switch is the easy part. Living with what it changes is the actual product.