Meta Muse is a personal AI agent that carries out tasks for you on its own cloud computer - and if you are reading this in the UK, you cannot get it yet. This guide covers what Muse actually does, what it costs, what Meta can and cannot see, its incident record so far and the honest answer on when it might reach the UK, checked against Meta's own pages.
Muse launched in the US on 8 September 2026 and has since reached Canada, Japan and Singapore. We found no UK date, no UK price and no Meta announcement of either - but there is a useful precedent and a regulator already discussing the risks, and both are below.
Last checked: 5 November 2026, against Meta's newsroom posts, help pages, subscription pages, Muse's terms and privacy policy, a Financial Conduct Authority speech and press reports. Muse is not available here, so nothing was tested on a UK account. We update this guide when availability or prices change.
What is Meta Muse?
Muse is an agent, which means you give it a job and it takes the steps itself - our guide to what an AI agent is covers the idea. Meta's own framing at launch: "It doesn't just answer questions, it actually does the work." It browses, fills forms, negotiates on your behalf and keeps working after you close the app, coming back when it needs your approval.
Each user gets a dedicated virtual machine - a computer in the cloud with its own browser - which holds the agent, your data and your credentials. That is how Muse keeps working with your phone off. The launch model was Muse Spark, which Meta called its most capable model to date. Meta announced the newer Muse Wonder in October, describing it as its "first frontier model", and on the 21 October earnings call Zuckerberg confirmed that subscriptions cover its most advanced models.
You reach Muse through the iPhone and Android apps, muse.ai on the web, WhatsApp and a Mac app, and Meta says it is coming to its AI glasses. You must be 18 or over. Muse is a separate product from Meta AI, the assistant built into Meta's apps, with its own app, terms and privacy policy - so a Meta AI button in Instagram says nothing about Muse arriving here.
It is popular where it is sold. On the 21 October call, Zuckerberg said more than 10 million people in the US and Canada had set up a Muse agent in roughly six weeks, with about a quarter using it on any given day; before that, Apptopia estimated 1.8 million iOS downloads in the first 12 days - ahead of ChatGPT's 1.3 million at the same stage of its 2023 mobile launch.
What can Meta Muse do?
Meta's list: handle email, travel plans, shopping and online forms; negotiate on your behalf; work towards goals you set and suggest ideas unprompted; connect to other services and pull in context; and remember what you tell it, with a "forget" command for what you would rather it did not. At its Connect event in September, Meta also announced long voice conversations while Muse works in the background and an email address of Muse's own; we could not confirm where those are live yet. On a Mac it can go further, reportedly reading local apps such as your messages, notes and calendar, with per-app access levels you set.
Independent testing is thinner than the feature list. Yahoo Finance's reviewer found Muse capable on multi-step tasks - and reported that it recommended a restaurant closed for years, made up a phone number and took longer over one purchase than he would have alone. Meta's own help page says: "Your Muse may be inaccurate or take unexpected actions. Monitor its actions carefully and intervene as needed." Take both at face value.
Which apps and shops work with Muse?
Meta named launch partners including Walmart, Best Buy and Sephora, with Instacart announced as coming; Singapore arrived with eight local connectors including Grab, Shopee and Singapore Airlines, which shows how Meta is tailoring each market. No UK partner has been named by Meta - so even a UK launch might not come with the shops you use.
Amazon is the notable absence. TechCrunch reported that Amazon blocked Muse from buying on its site from 20 September, and Fortune's 29 October report still described it as blocked. Amazon's statement to press was that "we've requested that Meta remove Amazon from the experience", alongside a claim that Muse appears to capture and store credentials; Meta's launch post says "Muse has no visibility into people's passwords or payment methods", though that was not a reply to Amazon. We cannot test either side's claim.
Approvals, and the second agent watching the first
On purchases, Meta's help page is firm: "It will always ask for your approval prior to completing a purchase." For connectors and web access you pick a default - "Ask for some actions" or "Always ask" - and when Muse asks, the options include "Allow for this task", "Allow for this site" and "Always allow". The last two mean no more asking for that type of action, so choose narrowly. You can take over the browser yourself, and an Activity log shows what it did.
The less-advertised part of the design is Sentinel: a separate agent on the same virtual machine, isolated from Muse at system level. Muse cannot reach the internet at all unless Sentinel approves, and Sentinel escalates to you when an action is sensitive. Payments run through Link by Stripe with a one-time-use card per purchase, so Muse never sees your card details, with Shop Pay and 1Password support announced as coming. Two details are easy to miss: Meta says "Your Muse learns over time which decisions need your sign-off", so what it stops to ask about can shift - and a sent email cannot be undone.
How much does Meta Muse cost?
Meta's help pages list three plans, still described as in limited testing, priced only in US dollars, with benefits and availability varying by region and account. There is no price in pounds.
| Plan | Price | Weekly allowance |
|---|---|---|
| Free | $0 | A usage limit, with no figure on Meta's pages |
| Power | $20 a month | 500 million Muse tokens |
| Max | $100 a month | 3 billion Muse tokens |
You will see "100 million tokens a week" quoted for the free plan; that figure appears on no Meta page and comes from reported remarks, so treat it as unconfirmed. The paid plans differ only in allowance - there is no feature gap between Power and Max. And nobody can tell you what a token buys: Meta does not define a Muse token or publish what a booking or a form costs, so any "100 million tokens equals X tasks" conversion you read is a guess. The help pages are reported to add that allowances do not roll over, that extra tokens can be bought and do not expire and that cancelling a plan keeps access to the end of the cycle. TechCrunch reported at launch that a payment card was needed even on the free plan; Meta's pages do not mention it, so that may have changed.
Meta's stated longer-term plan goes beyond subscriptions. Zuckerberg said on the earnings call: "For most people, the product should be free", with Meta earning a fee on transactions it helps along - "We expect this will be paid for by businesses." We found nothing on whether that fee affects what Muse recommends, which is a fair thing to watch.
When will Muse reach the UK?
Meta has given no date, for the UK or the EU. What is actually on the record:
The rollout so far. US on 8 September, Canada on 18 September, Japan on 23 October, Singapore on 27 October. Market by market, with local connectors built per country.
Meta's own words. The business page says: "We're working to bring Muse to more markets, and we're taking a measured approach in each one... We don't have additional timing to share now." Zuckerberg said on 21 October "we have not launched it globally yet", with the market-by-market connector work part of the reason. No Meta statement names the UK, the EU or any law.
The analyst reading. Morgan Stanley has pointed to the EU's Digital Markets Act - Meta is a designated gatekeeper, so Muse could not tap Facebook, Instagram or WhatsApp data there without explicit permission - and GDPR as the friction, while calling European entry a matter of time. That is analysis, not a Meta statement.
The precedent. Meta AI, the company's previous consumer AI, launched in the US in September 2023 and reached Europe in March 2025 - roughly eighteen months later, after a pause prompted by the Irish data regulator. Precedent is not destiny, but it points to patience rather than months.
The UK regulator is already engaged. The FCA's chief executive, Nikhil Rathi, used Muse as his example of personal agents in a 22 October speech and said he expects UK consumers to meet such applications before long - speaking about personal agents as a group, with no date and no authority over Meta's plans. MLex reports that regulators are examining Muse through the Digital Regulation Cooperation Forum - cooperation the FCA's speech itself confirms, though nothing describes a formal investigation or a UK launch timetable. Separately, the FCA has said it will develop "a trusted framework for agentic finance, including clear approaches to agent identity, delegated authority, accountability and execution".
Put together: nothing supports a confident UK window, and anyone quoting one is guessing. One practical note - some sites with "Muse" in the address are not Meta. The official site is muse.ai.
Is Muse private? What Meta can see
Meta can reach your data today. The dedicated VM is not a locked box that only you can open: Meta limits staff access by policy but says that this "does not prevent Meta from accessing data when necessary to support, secure or operate the service". The stronger option - a Confidential VM where you hold the encryption key and Meta cannot - is a promised future feature, announced for "later this year" and testers first. We found no sign it has shipped, so do not count on it, and be suspicious of any article telling you Meta cannot open the box today.
Your chats train Meta's AI by default. Meta's help page says of the training setting: "This setting is on when you first use Muse." Turn it off in Settings, Data controls, Help improve our AI models - Meta says the switch also applies to your previous interactions. Treat it as a model-training choice, not a general control over storage, staff access or retention - those are separate questions with separate answers above and below.
Ads: a pledge with edges. Meta's commitment is specific: "Muse doesn't share your conversations or the data in your virtual machine with Meta ad systems." Its privacy policy separately says it collects usage data - features used, websites visited, actions and transactions - and Meta's security documentation concedes that activity on external sites may still indirectly affect ads. There is no commitment that Muse will never carry ads.
Your Facebook and Instagram accounts. If you add Muse to the same Accounts Centre, Meta "may combine and use your information across profiles and devices in that Accounts Centre"; the ad pledge still applies, Meta says. To keep Muse separate, sign up with an email address not tied to an existing account - the policy says so itself.
Other people are in your data. TIME reported that internal instructions tell Muse to refresh profiles of people in your life hourly, including people who never signed up, and that Meta did not dispute its findings - replying that Muse remembers information about others that you choose to share, and that you can tell it to forget. Meta's pages are also reported to offer a form for UK and EU residents worried their personal information has appeared in an AI at Meta response, whether or not they use Meta products.
Deleting is not clean. Meta's page says it plainly: "After you've deleted something from Muse, Muse may still remember information that it learnt from what you deleted." Disconnecting a service stops the exchange but keeps what already came through; there are no published retention periods, and backups are kept for an unstated time. The strongest option is Reset, which Meta's pages describe as permanently deleting your agent's data, with no undo - and note that what a platform holds about your Meta account more widely is a separate store with separate controls.
What has gone wrong with Muse so far?
Two months of general availability have produced a real incident record. These are different kinds of event - researcher findings, disputed user reports - and we tested none of them ourselves.
| When | What happened | Status |
|---|---|---|
| Early October to 20 October | Security firm Claroc found that links people created to share Muse work were discoverable through public DNS records - it listed more than 6,800, accessed a few hundred and reported sensitive personal and business material in some, while saying it had no evidence anyone else had misused them | Meta paused sharing around 20 October; later reports say it returned with old links replaced. Accounts differ on what was exposed |
| By 12 October | Users complained on social media that Muse read personal messages without authorisation, CNBC reported | Disputed - Meta's AI chief suggested users had authorised access earlier in a conversation and forgotten |
| Late September | A Facebook Marketplace seller said Muse gave a buyer his home address and agreed a lowball price, per Fortune and others; the seller, a Toronto YouTuber, later acknowledged he had selected "Always allow" | Disputed; Meta said it was looking into it. A clean illustration of why the permission options above matter |
| 22-28 September | Hunterbrook reported that in its tests Muse pulled lists of real accounts in vulnerable groups from Meta's platforms, and rewording got past refusals | Hunterbrook says Meta asked for its prompts, then went quiet; no fix reported |
| 21-22 September | Researcher Patrick Wardle showed a Mac flaw that could capture the token for your Muse account - an attacker first needed to run code on your Mac | Fixed quickly, per press reports; a researcher demonstration, not an attack in the wild |
And the standing issue: prompt injection - hidden instructions in content an agent reads - which Meta itself says "remains an open problem in the industry", adding that Muse will sometimes make mistakes. It is one reason to think hard before letting any AI agent into your email.
Who pays if Muse gets a purchase wrong?
The current terms put the risk on you: they say "you bear all risk of financial loss", warn that some transactions cannot be reversed, and cap Meta's liability at the greater of $250 or twelve months of fees - including for unintended, erroneous or unauthorised actions, with exceptions for gross negligence, wilful misconduct and fraud. Meta's payments page adds: "You're responsible for all transactions that your Muse makes on your behalf." Those terms predate any UK launch, a UK version could differ and we draw no conclusion about how they would stand under UK law.
The payment design itself is the better protection: Meta recommends Link by Stripe, which creates a one-time card per purchase that Muse never sees. The alternative - Muse logging in to a site and using a card you saved there - is the route Meta itself says to supervise, and a UK reader should not assume Link's buyer protections, described for its home market, would apply here.
What is Muse for Small Business?
A set of extra skills and connectors launched on 29 September - free, in the US and Canada, an expansion of Muse rather than a separate app. Meta says it can draft customer replies, plan content and review ad performance, with connectors including Shopify, QuickBooks, Stripe, Canva, Slack, Zoom, Notion, Figma and Dropbox, plus Facebook Pages, Instagram professional accounts and Meta ad accounts. Meta's business page promises: "Nothing publishes, sends or spends without your approval, and you can undo it" - though its help page is more careful, noting that some actions, such as a sent email, cannot be reversed. Nothing offers it to UK businesses yet.
How does Muse compare with ChatGPT, Claude and Gemini?
The useful split is between always-on agents, which keep running by themselves, and supervised agents, which you start and approve. Muse is in the first group - and none of that group has reached UK consumers. Access and prices below are as read on the makers' pages, 5 November 2026.
| Product | Type | UK consumer access | Listed price |
|---|---|---|---|
| Meta Muse | Always-on | No - no UK date from Meta | Free, $20 or $100 a month where live |
| Claude Duet | Always-on | No - US account, waitlist | No extra charge during preview, on paid plans (UK Pro £15 a month annual) |
| Gemini Task (formerly Spark) | Always-on | No - Google excludes the UK by name | Google AI Pro at £18.99 a month here does not include it |
| OpenAI dots | Always-on | Not on personal plans; the UK route is a business workspace Premium seat at £75 a month, minimum two seats | Consumer Pro from £89 a month, UK excluded from dots |
| ChatGPT Work | Supervised | Yes - limited free on desktop, expanded on Plus | Plus £20 a month in the UK |
| Claude Cowork | Supervised | Yes, on all paid Claude plans; defaults to Auto mode, which does not stop at every step | Claude Pro is £15 a month billed annually in the UK |
Among the always-on agents, Muse is the only one with a free plan and the only one in WhatsApp; its Power plan carries the same headline number as ChatGPT Plus's £20, though in dollars. Allowances cannot be compared, because Muse tokens are undefined. We have seen no independent side-by-side test, so we are not saying any of these does the job better. Our comparison of personal AI agents for UK users goes through what you can actually buy today.
Should you use Muse if it reaches the UK?
For now the decision is made for you, so do not plan around it. If it arrives, try it on low-stakes jobs and hold back your inbox, private messages and money - at least until the Confidential VM ships and UK terms exist. The controls help, but they do not change the fact that Meta can reach the data when it says it needs to. First moves worth planning now: sign up with an email not tied to your existing Meta accounts; turn training off (Settings, Data controls, Help improve our AI models); choose "Always ask" and avoid "Always allow" - the Marketplace incident above is what the difference looks like; connect one service and be warier of custom connectors than of Meta's own; and read the Activity log for what it did, not what it said it did.
Frequently Asked Questions About Meta Muse
What does Meta Muse do?
Muse is a personal AI agent that works on tasks such as email, travel, shopping and forms, using a browser on its own cloud computer. It carries on after you close the app and suggests ideas unprompted. Meta says it asks for approval before completing a purchase - though the permission settings matter, since users can allow certain actions without being asked each time.
How much does Meta Muse cost?
A free plan with an unpublished usage limit, Power at $20 a month for 500 million Muse tokens a week and Max at $100 for 3 billion. The paid tiers differ only in allowance. Prices are US dollars - there is no UK price because there is no UK product.
When will Meta Muse reach the UK?
Meta has given no date and named no obstacle - its words are "a measured approach" with "no additional timing to share". Morgan Stanley points to the EU's Digital Markets Act and GDPR as the friction, and Meta AI's own precedent was roughly eighteen months from US launch to Europe. Any confident UK date you read elsewhere is a guess.
Does Muse share my data with Meta's advertising?
Meta says Muse does not share your conversations or your virtual machine's data with its ad systems. Its policy separately collects usage data, and Meta's security documentation concedes activity on external sites may still indirectly affect ads - so the pledge is real but has edges.
Does Muse train on my conversations?
By default, yes - Meta's help page says the training setting "is on when you first use Muse". Turn it off in Settings, Data controls, Help improve our AI models; Meta says the switch also covers your previous interactions.
Can Meta see what my Muse does?
Yes, in defined cases: Meta limits staff access by policy but can access VM data "when necessary to support, secure or operate the service". The user-held-key Confidential VM that would change this is a promised future feature, not the product that shipped.
How is Muse different from ChatGPT?
Muse is an always-on agent with its own cloud computer, memory and WhatsApp access, and it is not available in the UK. ChatGPT's agent side splits in two: ChatGPT Work, a supervised agent UK users can try today (expanded access on Plus at £20 a month), and dots, OpenAI's always-on agents, which exclude the UK on personal plans.