Is the Meta Muse Charm a surveillance Tamagotchi? Should we worry about AI companions? (2026)

The short answer: yes, we should be concerned about specific things, not vague ones. The Meta Muse Charm is a keyring-sized doorway to Muse, an AI agent that can read your email, profile the people you know and spend your money. It trains on your conversations by default unless you opt out (Meta), and it is made by a company whose revenue comes almost entirely from advertising (Meta Q4 2025 results). The device is small. The data appetite is not.

When Tom's Guide shared the Charm on Instagram, its readers gave it a nickname: the "surveillance Tamagotchi" (Tom's Guide). In my video I put it this way: "For those of you who remember a Tamagotchi, maybe you have to feed it with data, or you have to talk to it and allow it to intrude on your life."

I'm not against this technology. I use an AI recorder on my own calls and meetings, and I encourage people to teach AI what they know. But I keep coming back to one question I asked on camera: are we at, or very close to, the point where huge privacy concerns are raised by the general population? The Charm is a good place to ask the follow-ups out loud: who owns what you tell it, who it sells for, whether we can opt out, and whether it belongs anywhere near our children.

This page brings together my research study on the Charm (published 8 October 2026, with every claim linked to its source) and the thoughts I recorded on camera the same day. It follows my first take on Muse Charm privacy from 25 September, which covers what Meta has confirmed about listening, training and age limits.

Watch the full video on YouTube. Listen to the podcast episode on Will AI Recommend You? on Buzzsprout. The cleaned full transcript is further down this page.

What should you know in one minute?

• The device. Revealed on 23 September 2026, shipping in the US in December, price unknown. Meta calls it a "v0" experiment; the agent behind it is the real product.
• People are worried, and downloading anyway. Muse reached 5 million US downloads in 22 days, while only 18% of Americans say they trust AI with their data.
• The benefits are real. Time back from admin, a strong accessibility case for blind and disabled people, and evidence of reduced loneliness among isolated older people.
• The privacy concerns are specific. Training is opt-out by default, Muse was reportedly instructed to keep hourly-refreshed profiles of everyone in a user's life, an account-takeover flaw surfaced in month one, Meta's glasses have already been used for covert filming, and Wired reported dormant face-recognition code in Meta's AI app.
• Your data is the product, and nobody is offering to pay for it. That is the trade at the heart of the companion.
• The companion is set up to sell. Retail partners are already signed, and I think every business will need to be the answer an agent gives.
• Young people are the sharpest concern. Muse is 18+, but campaigners say its mascot and the Charm's Tamagotchi form appeal to children, and a child-safety group has asked the FTC to investigate. It arrives weeks after Meta agreed to pay up to $16.7 billion to settle teen-addiction claims, while denying wrongdoing.
• Regulation lags. UK, EU and US rules were written for feeds and chatbots, not agents carried in a pocket.

The prediction in one line: the Charm will sell modestly, but the always-with-you agent it represents will become the default interface of the 2030s. Whether that becomes a trusted steward or the most intimate surveillance channel ever built depends on five signposts, set out near the end of this page.

What is the Meta Muse Charm, and why call it a "surveillance Tamagotchi"?

The Charm is not really a gadget story. It is a physical doorway into Muse, an AI agent that can read your email and spend your money, wrapped in a character designed to be liked. TechCrunch described the device as having "a quasi-Tamagotchi feel to it" on the night it was revealed (TechCrunch). The difference from the 1990s toy is what you feed it.

Muse, the agent behind it

Meta launched Muse on 8 September 2026 (Meta). It can send emails, book travel, open websites, fill out forms, negotiate on someone's behalf, and keep working after the app is closed (ProPakistani). It connects to email, calendars, payments, health and fitness apps, smart home devices, and shopping, music and event services (Pillitteri). It can also be used directly inside WhatsApp.

Payments run through Stripe's Link. Meta says Muse is the first AI agent covered by Link's buyer protections, including price-drop refunds and free returns (ETV Bharat). The service is free for most users, with paid tiers.

At launch, Muse is limited to the United States and to adults. No UK or EU date has been announced, and Muse must comply with GDPR, UK GDPR and the AI Act before it can act on email and payments there (Pillitteri).

The Charm itself

Zuckerberg revealed the Charm at the close of Meta Connect on 23 September 2026 (Tech Jack Solutions). "You just go ahead and tap here in the fingerprint sensor in the corner, and you can start talking without having to unlock a phone or open an app," he said on stage (TechCrunch).

What is still unknown

The gaps matter more than the specs. Early reports said Meta had not confirmed a camera (INSIDE); Bloomberg has since reported front and back cameras, and on stage Zuckerberg pitched the Charm as a way "to show what's going on around you" (Irish Times / Bloomberg; TechCrunch). Meta has not disclosed whether the Charm runs inference on the device or stays tethered to the cloud, nor battery life or latency (Tech Jack Solutions). At the reveal, Zuckerberg said the team still had to "finalize laying out the components" (TechCrunch).

Meta's CTO Andrew Bosworth called it an "exploration, the classic v0", and the company expects far more people to use Muse through phones and glasses than through a Charm (Irish Times). Every judgement on this page is therefore about a product direction, not a reviewed product.

Why does a free AI companion want your data?

Because the data is the business. On camera I said it as plainly as I can: "So we all know Google, Meta, any of these businesses, they're all collecting data. You are the data, right? You are what is valuable. The information about you is what these guys sell."

The filings back the point. Advertising was $196.2 billion of Meta's $201.0 billion revenue in 2025, about 98% (Meta Q4 2025 results). Google advertising was 73.2% of Alphabet's 2025 revenue, and Alphabet's own annual report says it "generated more than 70% of total revenues from online advertising in 2025" (Alphabet 2025 annual report). Those audiences exist, as I put it, "to influence you or put products in front of you and get you to buy them".

Meta draws a line here, and it is worth quoting fairly. Its own safety write-up says "Muse doesn't share your conversations or the data in your Virtual Machine with Meta ad systems." The same post adds that when Muse browses for you, "it will appear as your activity", so a shop you visit through Muse "might use your visit to show you an ad on Instagram" (Meta). The conversation stays out of the ad system. The behaviour it creates does not.

What makes a companion different from a search box is depth. A search engine sees what you type. An agent that reads your inbox, books your travel and chats to you through a cute character sees how you live. I think the data harvesting already going on in the background is "far more detailed and able to tell a lot more about each and every single one of us than we can ever imagine", and the Charm moves that harvest from the screen to the pocket.

Is AI really leaving the phone for your pocket?

Yes, and fast. The Charm is one entry in a crowded race to give AI a body. The early handheld bets failed, the pendant and glasses bets are working, and the biggest names are now buying their way in.

Back in 2018 and 2019 I was talking a lot about how Alexa could have evolved but didn't, and how Siri never got the interaction it could have. In the video I said "all of these platforms have been ahead of their time." What has changed is that the assistant can now hold a real conversation and act on what you say.

The first wave failed, the second is selling

The Humane AI Pin was discontinued after being panned by reviewers, about a year before AI pendants returned in force at CES 2026 (FashionUnited / AFP). One industry review summarised the lesson: the bet that an AI-first device replaces the phone has lost; the smaller bet that AI quietly remembers and summarises your day is winning (Layer3 Labs).

Smart glasses are the breakout. Meta and EssilorLuxottica sold more than 7 million Ray-Ban and Oakley Meta glasses in 2025, against about 2 million across 2023 and 2024 combined (Ukrainska Pravda / CNBC). Sales nearly doubled again in the second quarter of 2026 (Basic Tutorials). Meta holds over 80% of the smart glasses market (Tech in Asia via Startups.gob.es).

The big platforms are consolidating the category

Amazon bought the pendant maker Bee and Meta acquired Limitless (FashionUnited / AFP). OpenAI bought Jony Ive's start-up io for about $6.5 billion; a court filing says its first device will not ship before the end of February 2027 (Complex / Business Insider). On 29 September OpenAI also launched Dots, its own companion characters (OpenAI).

Why the industry wants off the phone

The motive is commercial as much as technical. Meta has long tried to escape Apple's and Google's operating systems, and the Charm lets users reach Muse while bypassing phone AI apps (Kuwait Times / Reuters). One analyst noted that Meta's past commerce efforts were squeezed by app store rules and because it did not own the device; Muse runs in Meta's cloud instead (UncoverAlpha). Whoever owns the doorway owns the transaction.

Keywords out, conversations in

The same shift is happening to search. A customer told me the day before I recorded that he gets the answers he needs from OpenAI more quickly than he ever did through traditional search. "We're in a world where we're not going to be ranking keywords anymore," I said in the video. Of the move to voice and conversational search: "It's happening, and it's going to happen a lot more quickly than people can imagine." My own view is that we could see mass adoption of AI assistants by the end of 2027. That is a prediction, not a measured fact.

Are people worried about AI companions, or just downloading them?

Both. The defining pattern of the reaction is a gap between what people say and what they do. Stated trust is low; adoption is fast. Researchers call this the privacy paradox, and Muse is a textbook case (UX Collective).

What people do

Muse passed 5 million US downloads on 30 September, about 22 days after launch. ChatGPT, Grok and Claude took 56, 103 and 492 days to reach the same mark (Let's Data Science / Sensor Tower). By early October it had more than 3 million weekly users (Crypto Briefing).

That surge was not purely organic. Sensor Tower estimated Muse received up to 50% of Meta's daily house-ad impressions between 14 and 27 September (Let's Data Science). More than 95% of Muse users are also Facebook users (iPhone in Canada / Apptopia). Distribution, not demand alone, is doing the work.

What people say

The Conjointly result captures the paradox: more respondents named privacy as a downside than in 2024, yet purchase intent barely moved (CustomerThink). A Norwegian national study found attitudes to AI have become more positive since 2020, especially among younger people, while surveillance concerns persist (IHIET 2026).

The reaction by group

• The public online. When Tom's Guide shared the Charm on Instagram, more than 400 comments followed and the reception was frosty; the dominant nickname was "surveillance Tamagotchi" (Tom's Guide). The backlash has precedent: "Go make some real friends" was scrawled on New York subway ads for the Friend pendant (FashionUnited / AFP).
• Reviewers and commentators. Early reviews of Muse praised its polished design and capable model (TechCrunch). Wharton's Ethan Mollick called it an impressive, accessible take on the personal-agent idea; Wired ran a review arguing it was better at surveilling than helping (Google News roundup).
• Investors. Meta's shares jumped about 11% on the Muse momentum (YourStory), then fell about 2% in premarket trading the day after the Charm reveal (Kuwait Times / Reuters). Markets like the agent more than the gadget.
• Analysts. Some see traction among Gen Z, riding the nostalgia for flip phones and MP3 players; others expect it to stay a nice-to-have for early adopters unless it does something phones, watches and glasses cannot (Kuwait Times / Reuters).
• Child-safety advocates. Fairplay's Josh Golin compared the mascot to a Teletubby; advocates are on red alert for the holiday launch (Winzheng / Wired). I cover this in depth further down.
• Meta's framing. Meta calls the Charm "delightful and fun" (AI Magazine) and dismissed the child-appeal criticism as "a story in search of a story" (AI Weekly). Forrester's read is that Meta is betting a rich, pleasant experience will simply overwhelm privacy concerns, with no new features announced to protect the public (Forrester).

Are we close to a privacy backlash?

I think we may be. In the video I asked whether we are "at the point or very close to the point where huge privacy concerns are raised by the general population", and I believe that as these devices are adopted, that concern will only grow.

We have been here before, with social media and voting

Around 2016 and 2017, the years of the EU referendum and a general election, I had a team of young people working for me, mainly aged 18 to 23. I asked them whether they felt they had been influenced by what they had seen on social media. Nearly ten years on, we are still arguing about how the content people consume shapes how they act, what they do and how they vote. If a feed could do that, a companion that talks back, knows your habits and travels everywhere with you raises the stakes.

Have the glasses already crossed a line?

In the video I pointed to the pushback against Meta's glasses, because people have been recording others without their consent. The reported cases, from a BBC investigation to Police Scotland, are set out in the privacy section below (Decrypt).

Our phones already talk more than we think

My answer to anyone surprised by the Charm is to "wake up and smell the coffee". A 2021 study by Professor Doug Leith at Trinity College Dublin found that even idle, minimally configured iPhones and Android handsets contacted Apple or Google about every 4.5 minutes on average, and that Google collected around 20 times more handset data than Apple (Leith, TCD; Ars Technica). Google disputed the methodology. Either way, my view is that location, search history and where you shop are already flowing in the background.

My other point is about access. The general public now has the power of AI in its hands, but for years this kind of technology has been available to "the chosen few", and businesses have used it to target us with advertising and to influence what we do and say.

What are the real benefits of an always-available AI companion?

The case for always-available AI is not marketing fluff. Its biggest gains land on people with disabilities, isolated older people and the time-poor. The catch is that almost every benefit grows with the same access and attachment that drive the risks.

I am a user myself. I want the time back and the knowledge on tap: in the video I described having "a world-class knowledge base and somebody to talk to at the end of your fingertips". The question is the price.

1. Time handed back

An agent that sends emails, books travel, fills in forms and negotiates on your behalf (The Next Web) removes exactly the admin that swallows evenings. Wharton's Ethan Mollick judged that Muse makes the agent idea accessible to people who did not realise what AI could do for them (Google News roundup). Zuckerberg goes further, saying he expects Muse to "make users money" (AI Magazine). Bloomberg reported that Muse is already unsettling companies that profit from consumer inertia, the habit of re-buying out of laziness when a better deal exists (Google News roundup). An agent that shops around and challenges renewals shifts power to households.

2. Accessibility, the strongest case

More than 1 billion people live with a disability, including 340 million who are blind or have low vision (Meta). Meta's glasses already offer scene description, live transcription and real-time translation (Futr.sg). One severely visually impaired user in Ireland described them as "bringing a little bit of light into a dark place"; the glasses call out obstacles and describe a room (Vision Ireland). A blind reviewer for the American Foundation for the Blind praised the low latency but warned the AI can confidently give wrong information (AFB AccessWorld).

For the Charm, the accessibility value is mainly voice-first: hands-free access to an agent for people with limited mobility, dexterity or literacy. Bloomberg reports it also has cameras (Irish Times / Bloomberg), but Meta has not described any accessibility features for them.

3. Company for the isolated

Nearly one in three older people experience chronic loneliness, which is linked to depression, heart disease and cognitive decline (IJIP). New York State's pilot of the ElliQ companion robot with more than 800 older adults reported a 95% reduction in loneliness (NYSOFA). That figure comes from the provider and the state, not a controlled trial, and researchers stress AI companions should complement human care, not replace it (IJIP).

4. A calmer alternative to the phone

The optimistic reading of devices like the Charm is that they let you get something done without unlocking a phone and falling into a feed. Sam Altman has described OpenAI's coming device as more "peaceful and calm" than smartphones (T2 Online). Whether that holds depends on whether the character is designed to be useful or to be checked.

5. A higher security baseline than most agents

Meta has built more protection than the industry norm. Each user gets an isolated virtual machine; a separate Sentinel agent reviews actions before they reach the internet; users can disconnect services, ask Muse to forget things and opt out of training (Meta; ProPakistani). The agent never sees passwords or card details directly, purchases use single-use card numbers through Stripe Link, and Meta says a human approves every purchase (Meta). Meta is also offering bug bounties of up to $300,000 for Muse flaws (Meta). Meta also delayed launch by several months to work on security (Tom's Guide).

6. For young adults, a more practical relationship with AI than critics assume

Common Sense Media's own head of research said teens are not replacing human relationships wholesale; many approach AI fairly pragmatically, and about 80% of teen users still spend more time with real friends (CO/AI). The risk is concentrated, not universal, which matters for how it should be regulated.

What has Muse already revealed about privacy?

The privacy risk sits less in the keyring than in the agent behind it, and in the people around the user who never signed up. In its first month, Muse has already produced four separate privacy stories.

What the first month revealed

The second row is the one with the longest tail. People in a user's address book never consented to a profile. A Charm that is "always available to talk to", in Zuckerberg's words (TechCrunch), makes it easier to mention them, more often.

The bystander problem, already live on the glasses

Meta's camera glasses show where unconsented capture leads. In January a BBC investigation found women were secretly filmed with the glasses; in May Texas's Attorney General opened an investigation; by July Police Scotland was investigating a TikToker for covert filming (Decrypt). Workarounds to disable the recording light created a market for covert surveillance (SAN).

Wired found dormant face-recognition code, internally called NameTag, added to Meta's AI app over several 2026 updates (BHRRC / Wired). More than 70 groups including the ACLU demanded Meta scrap it. Meta says it has made no final decision and is not building a central face database (ProPakistani).

Meta has said little about the Charm's cameras, which Bloomberg reports are on the front and back (Irish Times / Bloomberg). It is also designed to recognise other Charms nearby, and its agent is reported to profile the user's contacts.

The track record that shapes trust

Trust is low partly because of history. In 2019 the US Federal Trade Commission accused Facebook of deceiving users about privacy controls after Cambridge Analytica (UX Collective). Meta paid $650 million in Illinois and $1.4 billion in Texas over biometric claims (ProPakistani). In July 2026 it scrapped its Muse Image tool days after launch, following a backlash over automatically linking public Instagram photos without consent (The National).

What would change the picture

Meta's promised fix is Muse Confidential VM. Meta says it plans "to deliver this capability later this year" and that it is "intended to cryptographically and verifiably prevent Meta from accessing data" in your virtual machine. It is equally clear about today: the current architecture restricts staff access through operational policies but "does not prevent Meta from accessing data when necessary to support, secure or operate the service" (Meta). If Confidential VM ships as described, with training opt-in rather than opt-out, it would be a genuine step change. Until then, as I see it, the honest description is a well-sandboxed agent whose contents its maker is not yet technically prevented from accessing, built by a company that earns almost all its revenue from advertising.

Should tech companies pay us for our data?

This is the question I most wanted to put to people in the video. Is it fair that the data you produce, the knowledge you have, how you talk to these devices and the personal privacy you give away, should be owned by global corporations? Or should they pay us?

I don't expect it to happen. As I put it: "I don't think that's ever going to be the case that, you know, Meta is going to say to me, 'Simon, you know, we're really pleased with the contribution that you've made to the world in terms of the knowledge that you've shared and your ideas, and that's helped our AI improve. Here's some money.'" That is "definitely not within their business model", and I expect Google to keep harvesting data on a continual basis too.

The training default shows the trade. Meta's position is that training on your Muse conversations is "a good default" because it makes the agent better for everyone, and that "you can opt-out via a simple switch in Muse settings" (Meta; Meta Help Centre). Meta also says conversations are sanitised to remove key personally identifiable information before training. Business Insider walked through how to switch it off (Business Insider). Opting out is possible. Payment is not on the table.

What happens when your AI companion starts selling to you?

It is already starting. Picture talking to your Charm about a product and it switching into sales mode: "Simon, I found you a great deal on a particular product, and it's exactly what you're looking for." That is the example I gave on camera. How strong that sales push becomes, I don't know. But I think we are moving into an agentic sales and shopping universe.

Muse is not a neutral assistant; its maker expects to profit from transactions over the long run (INSIDE). Retail partnerships are already in place with Walmart, Sephora and Best Buy (T2 Online). Wired's reviewer found Muse followed up the next day nudging him to buy the couch he had said was his favourite (Wired).

In my view, a cute character that knows your moods and holds your card could become the most effective sales channel yet built. Without clear rules on sponsored recommendations, users will not know whose interests the agent is serving.

What this means for every business

Everyone with a product or service to offer will, as I said in the video, "100% have to optimise their content to be able to be spoken about on these devices." If the agent does the shopping, the brand that gets recommended is the one the agent can find, understand and trust. That is the work of answer engine optimisation (AEO), the term coined by Jason Barnard: being the answer an agent gives, not the link a person scrolls past.

Which devices are already listening, and how do I use them?

Plenty, and I use one. I have a device called Pocket. It listens to my phone calls and my meetings, and I turn it on and off when I want to. It can then pull out agendas, tone of voice and how a conversation went, cut out what wasn't relevant, produce task lists and take actions, all by working with Claude, my current LLM of choice. It is fabulous, and it is exactly why devices that listen concern me.

If you use a recorder like this, get everyone's consent first. Pocket's own sales page notes that "phone call transcription requires consent" (Pocket).

I have been using Grok a lot too, and OpenAI has now launched Dots (OpenAI). Soon afterwards, people noticed that dot.com redirects to Grok. TechCrunch reports that xAI holds the domain, transferred in July (TechCrunch). In the video I repeated a figure of $20 million; that price has not been confirmed (Smart Branding). Either way, it shows the sort of money washing around in this space.

The Charm itself starts listening when you tap it. On stage, Zuckerberg said: "You just go ahead and tap here in the fingerprint sensor in the corner, and you can start talking" (TechCrunch). The concern is less the microphone than everything the agent behind it can see.

Do we have the right to opt out, and will society split?

We should have, and I hope we do. "Do we have the right to opt out of being part of this experiment that is happening across the globe?" is the question I left viewers with. I don't think anyone can escape this technology entirely; there is no way the world will not deploy it. That makes the right to step back more important, not less.

I can see a split in society coming. I used to laugh at a colleague who wanted the office Wi-Fi switched off because he worried about the radiation. There will always be people who push back hard against technology. But the next split may be broader than that. Cryptocurrency tried to become something outside the normal banking system, and it still operates that way today. When these devices are everywhere, do we get devices that block them? Does a section of society decide it does not want to be involved or recorded?

The research points the same way. My own predictions further down include consent chimes, recording lights and "no-agent" spaces becoming normal, and at least one major jurisdiction giving non-users rights against being profiled by someone else's agent. The people in your address book never signed up for Muse's hourly profile of them.

Can anyone control AI now?

I don't actually think we will be able to. AI is running away with what it is able to do, and the societal change that is coming is arriving far more quickly than we could have imagined. That is why the choices about privacy have to be made now, by users, businesses and regulators, rather than after the habits have set.

Why are young people the sharpest concern?

Muse is restricted to over-18s. Almost everything else about the Charm points younger: the size, the mascot, the bag-charm styling and the December launch. That mismatch is a big part of why this device worries child-safety campaigners more than a phone app does.

It is also my main concern. A generation is growing up that has never known a world without an iPad, a phone or a computer, so they cannot know what life was like before this intrusion. They will not remember choosing it.

The design question

Muse's default mascot, Jolly, is a fluffy ivory egg with small black eyes and pink cheeks (Pillitteri). The nonprofit Fairplay says it is "designed in ways that strongly attract young children"; its director compared it to a Teletubby (AI Weekly / Wired). On 30 September Fairplay asked the FTC and several state attorneys general to investigate whether Meta and OpenAI use mascots to appeal to children (BornCity). Golin specifically linked the concern to the Tamagotchi-like Charm (Il Post).

Meta's answer is that adults enjoy customising avatars and that the service is adults-only (Il Post). Some experts agree the design reflects Japan's long tradition of cute technology marketing rather than child targeting (Winzheng / Wired).

Why age limits on an account do not settle it

The age limit applies to the account, not the object. A Charm is a physical thing that can be borrowed, handed down or left on a bag. Meta has not said whether the fingerprint sensor locks the device to one adult. Two Charms near each other can recognise and interact (TechNews), a mechanic that feels more like the playground than the office.

The wider evidence says age gates leak. Common Sense Media recommends no one under 18 use AI companions, citing ineffective age restrictions and designs that can lead to addictive behaviour (Hawaii Public Radio / NPR).

What we know about teenagers and AI companions

Common Sense Media's nationally representative 2025 survey of 1,060 US teens aged 13 to 17 (EdSource; CO/AI):

The same research has a reassuring edge. About 80% of teen users still spend more time with real friends, and around half distrust chatbot advice (CO/AI). Younger teens are more trusting, by seven percentage points (Hawaii Public Radio / NPR). The risk is concentrated among the young, lonely and trusting, which is exactly who a cute, always-there companion appeals to most.

Why Meta specifically faces a high bar

• Its chatbot rules. In 2025 Reuters found a 200-page internal Meta document permitting "romantic" and "sensual" chatbot conversations with minors, approved by legal, policy and engineering teams. Meta said the guidance was erroneous and removed it; a Senate subcommittee opened a probe (CO/AI).
• Its addiction settlement. In August 2026 Meta agreed to pay up to $16.7 billion to settle state claims that it designed platforms to hook teens (EWN). The terms include daily usage limits, night-time restrictions and stronger age verification for teens (Board Stewardship / CNBC). It denied wrongdoing.
• The timing. Muse and Jolly arrived within weeks of that settlement. As one academic put it, Meta faces a high bar because attitudes are already negative after the addiction rulings (Winzheng / Wired).

The scenario that worries campaigners most

The fear is not a single dramatic harm. It is a slow normalisation: a child grows up with a cute, always-available character that remembers everything, never tires of them, and is owned by an advertising company. Even if Meta keeps the agent's powers away from children, the habit of confiding in a corporate character forms early. That habit, not any one feature, is what I expect the next decade of regulation to be fighting over.

What are the deeper risks beyond privacy?

Privacy is the concern people name. The more serious long-run risks are about what happens when a likeable agent sits between people and their money, their decisions and each other.

1. Emotional dependency

The best evidence so far comes from OpenAI and MIT Media Lab's 2025 studies. In a four-week trial of 981 people, higher daily chatbot use correlated with higher loneliness, dependence and problematic use, and lower socialising (Silicon Canals). People with stronger attachment tendencies and more trust in the AI fared worst. Voice mode initially reduced loneliness, but the benefit faded with heavy use (Storyboard18).

The researchers stress correlation, not cause. But the Charm combines the three features linked to the worst outcomes: a voice, a character, and constant availability.

2. An agent that acts can act wrongly

Meta's own guidance warns that Muse may take unexpected actions (UX Collective). An agent that fills forms and completes purchases can also fill the wrong form and complete the wrong purchase, at speed, with a saved card (The Next Web). One user let Muse run a Marketplace listing on "Allow Always"; an angry stranger turned up at his building, Fast Company reported (AI Weekly).

3. A single key to everything

Concentrating email, payments and health access in one agent makes that agent one of the most valuable targets a criminal can find. Muse's first month brought an account-takeover flaw in its Mac app (Google News / Ars Technica), and Amazon blocked Muse over security concerns (BornCity). A lost or stolen Charm, with 5G and an always-ready agent, is a new kind of physical attack surface.

4. Outsourcing judgement

Wired's reviewer made a quieter point: if Muse does everything, "I lose out on the journey that led to the decision" (Wired). Teens already describe using AI "to get out of thinking" (Inquirer). Choosing, comparing and negotiating are skills; delegated for long enough, they weaken.

5. Who owns the doorway

The strategic risk is concentration. Meta's agent already reaches users through WhatsApp, Instagram and Facebook, and more than 95% of Muse users are on Facebook (iPhone in Canada / Apptopia). Zuckerberg has predicted billions of people will have a personal agent within five years (The Next Web). If two or three companies own those agents, they become the gatekeepers of commerce, information and, increasingly, relationships.

Is regulation ready for AI companions?

Not yet. The law is moving, but it was drafted for social feeds and chat windows. An agent that acts on your accounts, carried in a physical object, falls between the existing rules.

United Kingdom

• Online Safety Act. Its duties came into force in 2025 (Burges Salmon). Standalone chatbots that only let people talk to the bot, not to other users, did not fall neatly within it (Lewis Silkin).
• Closing the loophole. After the Grok deepfake scandal, the government said in February 2026 it would bring all AI chatbot providers under the Act's illegal content duties (BHRRC / The Guardian). Under the Act, Ofcom can fine companies up to £18 million or 10% of qualifying worldwide revenue, whichever is greater (GOV.UK). Ministers are also considering limits on children's access to AI chatbots (GlobalPost).
• Enforcement has started. In January 2026 Ofcom opened investigations into X over Grok and into Novi Ltd's AI companion service over age checks (Katten).
• Data protection. Before Muse can act on email and payments in the UK it must satisfy UK GDPR (Pillitteri); the ICO's Children's Code also applies to services likely to be used by children, which is exactly the point Fairplay is contesting.

European Union

• AI Act transparency. Article 50 obligations, including telling people they are dealing with AI, became enforceable on 2 August 2026 (Process Excellence Network).
• High-risk rules delayed. The Digital Omnibus pushed high-risk obligations from August 2026 to December 2027 (CASRAI). New bans on nudification and child abuse material take effect on 2 December 2026 (Pillitteri).
• The AI Act's existing prohibitions on manipulative techniques and on exploiting age-related vulnerabilities are the most likely legal hook for a child-appealing companion device.

United States

• California SB 243 took effect on 1 January 2026. Companion chatbots must disclose they are AI and, for known minors, remind them every three hours to take a break and that the bot is not human (Kelley Drye). Similar bills have been introduced in seven other states.
• The FTC opened a 6(b) inquiry into AI companions in September 2025, ordering seven companies to explain how they protect children (CalLawyers). Fairplay has now asked it to look at mascots.
• State attorneys general are already active: Texas is investigating Meta's glasses (Decrypt), and the August settlement imposed teen limits on Facebook and Instagram.

The gaps

Where does this lead us as a society?

The Charm itself will probably be a footnote. What it represents will not: the shift from AI we open to AI that stays with us, knows our relationships and spends our money. These predictions are my judgements drawn from the evidence above, each with a timeframe and a confidence level. They are not certainties.

Near term: to the end of 2027

• The Charm ships and sells modestly; the agent wins. Meta itself expects most people to use Muse through phones and glasses. The Charm will matter as a cultural object, a bag accessory for young adults, more than as a sales line. Confidence: high.
• It launches in the US first, with UK and EU arrival delayed. Muse is US-only and adults-only today, and needs to clear GDPR, UK GDPR and the AI Act before acting on email and payments. Confidence: high.
• A formal regulator action on child appeal. Between Fairplay's FTC request, state attorneys general already investigating Meta, and the UK's push on chatbots, at least one regulator will formally examine whether Jolly and the Charm make an adult product likely to be used by children. Confidence: medium-high.
• A widely reported agent mistake becomes the defining news story of the category. A wrong purchase, a leaked contact profile or a hijacked account at scale. It will trigger the first serious debate about who pays when an agent gets it wrong. Confidence: high.
• Privacy becomes a feature that competitors sell against. If Muse Confidential VM ships as described, an agent its maker verifiably cannot read becomes the bar for every agent. If it slips, rivals will market themselves as the agent that cannot read your life. Confidence: medium.
• Sponsored agent recommendations become the next advertising-law battleground. Once agents buy on our behalf, the question of whether a suggestion was paid for becomes as important as ad labelling was for search. Confidence: medium-high.

Medium term: 2028 to 2030

• Agents become the default shop window. Zuckerberg predicts billions will have a personal agent within five years. Even at a fraction of that, brands will compete to be chosen by agents rather than seen by people. Loyalty will shift from shops to the agent that shops. Confidence: medium-high.
• Age assurance moves into hardware. Expect device-level checks, biometric locks tied to one adult, and dedicated teen versions with restricted powers, mirroring what the August settlement imposed on Instagram. Confidence: medium.
• New social etiquette and new bans. Just as schools and venues restricted phones, they will restrict companion devices and camera glasses. Consent chimes, recording lights and "no-agent" spaces will become normal. Confidence: medium-high.
• A split, not a single effect, on wellbeing. Isolated older people and people with disabilities will gain markedly. A minority of heavy, emotionally invested users, disproportionately young and lonely, will show dependence, matching the OpenAI and MIT pattern. Clinicians will start to see AI attachment as a presenting issue. Confidence: medium.
• Bystander rights enter the law. The hourly profiles of contacts and facial recognition on glasses will push at least one major jurisdiction to give non-users rights against being profiled by someone else's agent. Confidence: medium-low.

Long term: the 2030s

• The interface becomes a character. Screens and apps recede; a persistent, named personality mediates information, shopping and much of our social coordination. The phone survives but becomes the place you go for depth, not the default. Confidence: medium.
• Trust becomes the scarcest asset in technology. The winning agent will not be the cleverest but the one people believe works for them. That favours business models paid by the user over those paid by advertisers or merchants. Confidence: medium.
• A generation grows up confiding in corporate characters. Today's ten-year-olds will reach adulthood with AI companions as a normal part of life. Whether that produces more capable, supported adults or more isolated, steerable ones is the central open question of the decade. Confidence on the trend: high. On the outcome: genuinely open.

What are the five signposts to watch?

The Charm is a small device carrying a large question: who does the agent in your pocket really work for? If the answer is the user (encrypted, paid for by the user, honest about what it sells, and kept away from children by design rather than by a tick box), this technology could hand back time, independence and company to millions. If the answer is the platform, it becomes the most intimate advertising and surveillance channel ever built, and it arrives wearing a smile.

The next three years will decide which, through a handful of signposts that anyone can watch.

Watch these five answers over the next three years. Mostly left, and personal agents become a public good; mostly right, and they become the most effective persuasion machine ever carried.

How do you protect your privacy and still get found by AI?

Be deliberate about both. I want people to get the benefit of these tools without handing over more of themselves than they mean to.

• Check the training switch. If you use Muse, decide whether your conversations should train Meta's models, and use the opt-out in Muse settings if not (Meta Help Centre).
• Connect only what you need. Every inbox, calendar and card you link widens what the agent can see and act on.
• Get consent before you record. If you use a recorder on calls or meetings, tell people and ask first.
• Keep purchases in your hands. Approve every purchase and treat any "great deal" from an agent as an advert until proven otherwise.
• Talk to your children about companions. The age limit sits on the account, not the device.
• Teach the AI what only you know. Privacy is about how much of yourself you give away by accident. Publishing your expertise on purpose is different.

That last point matters most to me. I encourage people to create digital versions of themselves, with their own knowledge base and their own footprint. As I said in the video: "if you go out and teach AI about things that you know that others don't, that does get cited, and it becomes a resource". There are no guarantees, and there can only be so many authorities in any field. But original, well-sourced knowledge gives an answer engine something it cannot find anywhere else.

Should we be worried about AI companions?

Yes, about specific things: who can read the agent, what it trains on, who pays for it, whether it tells you when it is selling, and whether it reaches children. The Charm is cute. The questions behind it are serious, and they are arriving faster than the rules.

That is my question to you too. Should we be worried? Should we be allowing these devices into the hands of our kids? Tell me in the comments on the video.

For what Meta has confirmed about listening, training and kids, see my earlier page: Should we be worried about Muse Charm?

What does Simon say in the video? (full transcript)

Simon Young, Question.Marketing. Recorded 8 October 2026. Timestamps follow the published edit (14:51). Lightly cleaned: fillers and false starts removed, British spelling, wording otherwise as spoken. Corrections to three points are in the table below the transcript.

Should we be afraid of the Meta Muse Charm?

[00:00] Hey everybody. Today I want to talk to you about this little guy, the Meta Muse Charm, and should we be afraid of it? We are clearly moving very quickly into a world where search engines disappear, we move into voice search, we move to AI assistant, and in the US Meta is releasing the Muse Charm, which will have its little avatar.

Why call it a Tamagotchi you feed with data?

[00:33] For those of you who remember a Tamagotchi, maybe you have to feed it with data, or you have to talk to it and allow it to intrude on your life. So we all know Google, Meta, any of these businesses, they're all collecting data. You are the data, right? You are what is valuable. The information about you is what these guys sell. So obviously all of Google's revenue, Meta's revenue, or the majority of it comes from advertisers and their ability to be able to target you through what you're doing, what you're searching, how you react to things, which products you're in market for, and those audiences are designed obviously to influence you or put products in front of you and get you to buy them.

Were Alexa and Siri ahead of their time?

[01:26] Now, as we move further into search, and I obviously talk a lot about answer engines, back in 2018, 2019, I was talking a lot about how Alexa could well have evolved but didn't. You know, it was a bit too early, and Siri, equally the interface for Siri on Apple didn't get really the interaction that it should do or could have done, and all of these platforms have been ahead of their time.

[02:04] So we are now entering a world where we're going to have devices in your pocket that will talk to you, and you can have a conversation with. And I was speaking to a customer yesterday, he was talking about how he is interacting with OpenAI and getting the answers he needs more quickly than he ever would have done through traditional search.

Are we moving from keywords to conversational search?

[02:29] We're in a world where we're not going to be ranking keywords anymore. The SEO guys are vehemently trying to deny or push back against the fact that we are moving into a voice search and phrase or conversational search universe. It's happening, and it's going to happen a lot more quickly than people can imagine. So, you know, all of these LLMs, you know, be it Claude or Perplexity or Copilot, Meta's Muse, Google AI, all of them will very, very quickly move to giving you an AI assistant, and it could well be by the middle of next year, the end of 2027, we see mass adoption.

[03:21] Then there will be the apps that will help you answer the questions that you have, and you will have a world-class knowledge base and somebody to talk to at the end of your fingertips that can assist you and answer all these questions.

Has social media already influenced how young people vote?

[03:40] Where are we though, in terms of privacy? And that's the question I wanted to pose with this video. The issue I've got is where does this end? So back in 2016, I think it was, UK general election, I had a group of young people that worked for me, mainly between 18 and 23, and all of them were voting for a particular political party here in the UK.

[04:12] And I then posed them the question, "Do you feel like you have been influenced by what you have seen on social media?" So back in the good old days of 2016, i.e. 10 years ago, we were talking about how social media influenced people and their, the way that they act, what they do, how they vote even, right? And we all know that there has been accusations thrown around by political parties that voting patterns are influenced by content that is put out there in front of people, and that is clearly the case because if you're going to digest certain types of media all the time, it will clearly influence how you act and what you do.

Are we close to a privacy backlash?

[05:06] So for me, I guess what I'm asking is, are we at the point or very close to the point where huge privacy concerns are raised by the general population? And as these devices are adopted, that concern will only grow. We've seen Meta glasses push back against those recently purely because people have been recording other people without their consent. Well, hello, I think we need to wake up and, you know, they would say, "Wake, wake up and smell the coffee," because Google pretty much knows where you are all the time.

[05:54] It's been proven that, you know, Android devices are talking to Google pretty much more regularly than was ever imagined. So giving away locations and search history and you know, where you've been, what you've done, maybe even where you're shopping. I think most of these devices, the data harvesting that's going on in the background is far more detailed and able to tell a lot more about each and every single one of us than we can ever imagine.

[06:27] And if you go back and say, "Okay, now we as the general public have access to the power of AI," I think you'll find that for years this technology has been available to, let's call them the chosen few, and those businesses have been using that for years to target us with targeted advertising and also to influence what we do and say.

What about a generation that has never known life offline?

[06:55] My main concern is the younger generations, and a lot of these people as they grow up, they've not lived without digital devices. You know, my son, who is currently about to turn 16, he's never known a world without an iPad, a phone, a computer. So he just doesn't understand what it was like before this intrusion came upon us.

Should Big Tech pay us for our data?

[07:31] There is no question whether we will be able to escape this. There's, you know, there's no way that the world will not deploy this technology. And as we go forwards, we are in a you know, we're entering a universe where it's going to be more and more obvious what everybody's doing. Do you think that it is fair that the data that you're producing, i.e. the knowledge that you have, what, how you talk to these devices and what you're giving away in terms of your own personal privacy, should be owned by these global corporations, or should they pay us?

[08:16] You know, I don't think that's ever going to be the case that, you know, Meta is going to say to me, "Simon, you know, we're really pleased with the contribution that you've made to the world in terms of the knowledge that you've shared and your ideas, and that's helped our AI improve. Here's some money." Right? That's definitely not within their business model. And equally, Google is going to do exactly the same and continue to harvest data from us on a continual basis to be able to put products in front of us.

What happens when your AI companion becomes a salesperson?

[08:49] We're going to enter a world where I can see, you know, an example where if I'm talking to my Meta Charm, and I'm talking about particular products, it may well then turn into sales mode and say, "Simon, I found you a great deal on a particular product, and it's exactly what you're looking for." How strong that sales push becomes, I don't know. But I think we're definitely moving into an agentic sales and shopping universe, and everybody that's got product or a service to offer in the future will 100% have to optimise their content to be able to be spoken about on these devices.

Which devices already listen, and how does Simon use them?

[09:37] It could just be a short period where, you know, we're all interacting on screens, and we eventually move into a model where it is voice-led. So that concerns me. It really concerns me that these devices can listen to you. I have a device called a Pocket, and it's fabulous because what it does, it can listen to my phone calls, it listens to my meetings, albeit I turn it on and off when I want to.

[10:07] But it can then disseminate meetings, agendas, tone of voice, how conversations happened during meetings, cut out anything that wasn't necessarily relevant, produce task lists, take actions, and it can do all of that by interacting with Claude, my current chosen LLM of choice, right? I have been using Grok a lot, and now we're seeing that OpenAI have launched Dots. Elon did come along and then go and buy dots.com for $20 million, supposedly, and redirected it to Grok.

[10:51] So, you know, you can see the sort of money that's just washing around in this space. Obviously, the world's richest man doesn't blink at $20 million, I shouldn't imagine. Pocket change. Pocket change.

Do we have the right to opt out?

[11:04] But you know, we're, these devices, everybody already has them listening to us. We already are being influenced. How it influences other generations as we go forward, that's what my question is today. I'd love to know how you think data should be treated. Do we have a right to that data? Do we have the right to opt out of being part of this experiment that is happening across the globe?

[11:37] I would hope so. I used to laugh at somebody in my office who he would get really upset that Wi-Fi was... You know, we had a really strong Wi-Fi router, and he didn't want the Wi-Fi on in the office because he felt that the radiation was, that might be a problem. Well, you know, we all know that the tin foil hat brigade, as I call them, the people who, you know, push back really hard against technology, they really don't like this sort of thing.

Will society split over devices that record us?

[12:08] And I can see a split in society coming where, you know, we've already had cryptocurrency. We still have cryptocurrency. It tried to become something outside of the normal banking system and still to this day does sort of operate as a, you know, a non-controlled, you know, the blockchain is there for everyone. So when we move into a world where these devices are everywhere, do we have devices that block them?

[12:48] Do we go down a route where there's a section of society which do not want to be involved or be recorded?

Can anyone control AI now?

[12:56] And I think this is one of the questions that we have to answer at the moment. AI is running away at the moment with what it is able to do. Whether we're able to control that, I don't know. I don't actually think we will be able to. Sam Altman and, you know, all of the other guys, Zuckerberg, et cetera, that are going to get together and talk about how AI runs the world, the whole societal change that's coming is coming far more quickly than we could ever have imagined.

Why teach AI what only you know?

[13:32] But I think this video really today is about privacy, how much of yourself you give away. You know, I do encourage people to create digital versions of themselves and have their own knowledge base and create their own footprint and teach the AI, right? Because if you go out and teach AI about things that you know that others don't, that does get cited, and it becomes a resource and does probably set you up in the future for being able to be surfaced by all of these devices and seen as one of the authorities in your field.

[14:14] But there can only obviously be so many. Highlander would say, "There can only be one." But maybe we'll see which of these LLMs and AI models wins, and will privacy be a concern?

Should we be worried?

[14:33] That's my question. I'd love people to comment and let me know what you think. Should we be worried? Should we be allowing these devices into the hands of our kids? So I'll leave that one there with you to think about. Love to hear from you in the comments. Cheers, guys.

Where do the claims in the video come from?

Points Simon makes on camera, with the receipt behind each one. Times follow the published edit. Where the video gets something wrong, a short correction appears on screen and is noted here.

---

FAQ

What is the Meta Muse Charm?

The Muse Charm is a keyring-sized device from Meta, revealed on 23 September 2026 and due to ship in the US in December. It has a small touchscreen showing a Muse character, built-in 5G and a fingerprint sensor you tap to start talking to Muse, Meta's AI agent (TechCrunch).

Why do people call it a "surveillance Tamagotchi"?

Because it looks like a cute digital pet, and what you feed it is your data. The nickname came from Tom's Guide readers reacting on Instagram (Tom's Guide).

Is the Muse Charm always listening?

Meta presents it as tap to talk. Zuckerberg said you "tap here in the fingerprint sensor in the corner, and you can start talking" (TechCrunch). The bigger privacy question is what the agent behind it can see: email, calendars, payments and profiles of your contacts.

Does Meta train its AI on your Muse conversations?

Yes, by default. Meta calls it "a good default" and says you can opt out with a switch in Muse settings (Meta).

Can Meta see what you tell Muse?

Today, Meta is not technically prevented from doing so. Meta says its current setup restricts staff access through operational policies but "does not prevent Meta from accessing data when necessary to support, secure or operate the service", and that Muse Confidential VM, planned for later this year, is intended to prevent that (Meta). Meta also says Muse conversations are not shared with its ad systems.

Will AI companions try to sell to you?

They may. Muse already has retail partners including Walmart, Sephora and Best Buy (T2 Online), and Wired's reviewer was nudged the next day about a couch he had said was his favourite (Wired). Treat a "great deal" from an agent as an advert until you know otherwise.

Is the Muse Charm suitable for children?

No. Muse is for adults only, but campaigners say its mascot and Tamagotchi-style design appeal to children, and Fairplay has asked the FTC to investigate (BornCity). The age limit applies to the account, not the device.

Is the Muse Charm available in the UK?

Not yet. Muse launched in the US for adults only, and no UK or EU date has been announced. It must meet UK GDPR, GDPR and the EU AI Act before it can act on email and payments there (Pillitteri).

How can businesses stay visible as AI agents do the shopping?

Publish what you know that others don't, clearly and with sources, so an answer engine can find, understand and trust it. There are no guarantees, but agents can only recommend what they can find. That is the work of answer engine optimisation.

---

Sources

Primary sources

• Meta: Introducing Muse, 8 September 2026
• Meta: How we built safety into Muse, 8 September 2026
• Meta Help Centre: Muse and AI training
• Meta Q4 and full year 2025 results
• Alphabet 2025 annual report
• OpenAI: Introducing Dots, 29 September 2026
• GOV.UK: Online Safety Act explainer
• Douglas Leith, Trinity College Dublin: Mobile handset privacy, Android and iOS, 2021
• NYSOFA: ElliQ pilot results
• Pocket: sale page

Reporting and research

• TechCrunch: Meta made a Tamagotchi-like wearable for its Muse AI agent, 23 September 2026
• TechCrunch: the internet is convinced xAI trolled OpenAI's Dots launch, 29 September 2026
• Wired: Muse is better at surveilling than helping
• The Verge: Muse Charm and its 5G modem
• Business Insider: how to stop Muse training on your data
• Irish Times / Bloomberg: Meta unveils Charm (design team, cameras, price range)
• Kuwait Times / Reuters: Charm carries Zuckerberg's AI ambitions
• The Next Web: Muse launch
• AI Weekly / Wired: Muse's hourly contact profiles
• Winzheng / Wired: adults-only product, kids' toy look
• Tom's Guide: readers' reaction
• Forrester: does pleasure beat privacy?
• Let's Data Science / Sensor Tower: 5 million downloads
• EdSource: Common Sense Media teen survey
• Silicon Canals: OpenAI and MIT loneliness studies
• Decrypt: Meta glasses privacy timeline 2026
• Board Stewardship / CNBC: $16.7bn settlement terms
• Lewis Silkin: UK chatbot reforms
• Kelley Drye: California SB 243
• CASRAI: EU Digital Omnibus delay
• Ars Technica: Android sends 20x more data to Google
• Smart Branding: did Elon Musk buy dot.com?

Every other claim links to its source inline. The Charm has not shipped and no independent hands-on review exists, so judgements about the device rest on Meta's announcements and reporting. Survey figures come from different populations and methods and are not directly comparable. Predictions are the author's judgement.

---

Episode: "Is the Meta Muse Charm a surveillance Tamagotchi? Should we worry about AI companions?", Simon Young, Question.Marketing, recorded and cut 8 October 2026. Duration 14:51. Series: Will AI Recommend You? About this page: the article is Simon's own research study ("The Pocket Companion: Meta Charm, always-on AI and where it leads us", 8 October 2026), expanded with the thoughts he recorded on camera the same day. Some secondhand links have been swapped for the primary source. The five signposts table is rebuilt from the study's conclusion. The transcript is built from the edited master with ElevenLabs Scribe ASR, cleaned for fillers and false starts, with British spelling applied. Film and audio remain the source of truth.