Meta has used the name Muse for two different things since April 2026, and the overlap is confusing enough that most coverage glosses over it. One Muse is a model family built by Meta's research lab. The other Muse is a consumer app that books your travel and answers your email. They are related, but they are not the same product, and knowing which one a headline is actually talking about changes what the claim means.
The agent is the newer and more consequential of the two. Meta introduced it on September 8, 2026 as, in the company's own words, the world's first personal AI agent built for everyone, and it launched carrying the weight of Meta's long and genuinely rocky privacy history, which is the first question most coverage of it has asked.
This post separates the two Muses: what Muse Spark the model actually is, what Muse the agent does and how it connects to your apps, what it costs, the security claims Meta is making and why skepticism is warranted, and what has already changed since launch.
Muse Is Two Different Things, and That Is Where the Confusion Starts
Muse Spark is a model, built by a division Meta calls Meta Superintelligence Labs. Muse, without the Spark, is a personal AI agent product, an app that uses Muse Spark to actually do things on your behalf. The model is the engine. The agent is the car. Reporting on one does not always make clear which one it means, and Meta's own naming does not help.
What Is Muse Spark, the Model?
Meta announced Muse Spark on April 8, 2026, describing it as the first model from Meta Superintelligence Labs, a division Meta says spent nine months rebuilding its AI stack from the ground up. Meta's own framing is that Muse Spark is small and fast by design, yet capable enough to reason through complex questions in science, math, and health. It handles multimodal input, can write visual code for websites and simple games, deploys parallel subagents to work on different parts of a question at once, and supports real time voice conversation and live camera analysis.
Meta's announcement did not publish benchmark comparisons against other models, and it stopped short of claiming Muse Spark leads any public leaderboard. It said only that the model currently powers the Meta AI app and that larger models were already in development. The current version, Muse Spark 1.3, shipped on September 2, 2026, and is the version that now runs the Muse agent.
What Is Muse, the Personal AI Agent?
Muse the agent launched on September 8, 2026, rolling out first in the United States on iOS, Android, and the web at muse.ai, with WhatsApp access and a future integration with Meta's AI glasses. Meta's own description is direct about the goal: an agent that sends emails, books travel, fills out forms, negotiates on your behalf, builds personalized plans toward longer term goals, and keeps working after you close the app.
It is built to be proactive rather than purely reactive. Meta says Muse remembers your preferences over time and surfaces suggestions on its own, and that it can turn an Instagram recipe reel into a grocery list or put together a party invitation without being walked through each step.
How Does Muse Actually Connect to Your Apps?
You choose which apps Muse connects to, and for sensitive ones, you choose how much access it gets. For email specifically, Meta says you decide whether Muse can only read your messages or is also allowed to send on your behalf. Beyond the apps Meta has built direct connectors for, Muse can reach other services through public APIs or, when no API exists, by operating a browser the way a person would.
That per-app, per-permission model is the main lever you have over what Muse is actually allowed to touch, and it is worth treating deliberately rather than granting broad access by default.
What Does Muse Cost?
Meta's own announcement is vague on exact pricing, saying only that Muse is free for most of what people need, with subscription plans for people who want to do more. Independent reporting since launch has been more specific: a free tier with a weekly token allowance, a Power tier at $20 a month with a larger allowance, and a Maximum tier at $100 a month with the largest allowance for heavy task delegation. A payment card is required to start even on the free tier, though Meta has said it expects most users to stay there.
What Security Does Meta Claim, and Should You Trust It?
Meta says Muse runs inside an isolated environment it calls Muse Secure VM, built so that, in the company's words, no one else's agent can reach it. The stated protections include credential encryption, an internal approval layer Meta calls Sentinel, full audit trails of what the agent did, an option to opt your data out of model training, and a promise that your conversations with Muse are not shared with Meta's ad systems. Meta has also said a future Muse Confidential VM will encrypt data with keys only the user holds.
Those are real, specific claims, and they deserve to be weighed against Meta's actual track record rather than taken at face value. The company settled with the FTC in 2019 for $5 billion over privacy violations, following an earlier 2011 settlement for deceiving users about privacy, and the Cambridge Analytica scandal remains the reference point for what can go wrong when a platform's data practices outrun its promises. In 2023 Meta was charged with violating an existing FTC privacy order, and the company recently settled an $18 billion multistate lawsuit over social media's harm to children. None of that proves Muse Secure VM will fail, but it is the honest context any claim about a new Meta security system has to sit inside, and TechCrunch's framing of the launch, asking plainly whether consumers will trust it, reflects that history rather than inventing a new concern.
What's New: The Updates Announced Since Launch
Meta has kept shipping on Muse quickly. Updates announced in the weeks after launch include a Muse Realtime Avatar for live video conversations with your agent, Mac desktop control so Muse can operate applications while you are away, a dedicated email address for Muse so you can loop it into a thread directly, and new integration partners including Stripe, Shopify, Best Buy, Gap, Sephora, Walmart, Wayfair, Expedia, Instacart, PayPal, GitHub, Granola, and Notion. Meta also opened a developer program for third-party integrations, which it said drew more than 1,500 applications in under a week.
How Does This Compare to What Others Are Building?
Muse the agent is not arriving in a vacuum. It lands in the same stretch of 2026 as OpenAI positioning GPT-6 Astra as a computer use model and the broader shift away from standalone AI browsers like ChatGPT Atlas toward agentic capability built into the core product. Every major lab is converging on the same bet: an assistant that acts across your apps and accounts rather than one that only answers questions. The products differ in the details, but the underlying direction, proactive, account connected, running in the background, is now the industry's default ambition rather than any one company's experiment.
Where This Leaves Things
Muse Spark is a real, if modestly documented, model that now underpins a consumer product with genuine reach. Muse the agent is a serious attempt at a category, personal agents that act on your accounts rather than just chat with you, and Meta's security claims are specific enough to evaluate rather than dismiss outright. Whether that is enough to outweigh the company's history is a judgment each user has to make with their own data, not something a feature list can settle for them.
If you are trying to figure out which of these new agent products are worth adopting for your own work, and how to do it without handing over more access than you intend to, our AI training service is built to help your team make that call carefully. Contact us if you want a second opinion before you connect anything.
Frequently asked questions
What is Meta Muse?
Meta Muse refers to two related things: Muse Spark, a model family built by Meta Superintelligence Labs and first announced in April 2026, and Muse, a personal AI agent app launched September 8, 2026 that uses Muse Spark to perform tasks like sending email, booking travel, and filling out forms on a user's behalf.
Is Muse Spark the same as the Muse app?
No. Muse Spark is the underlying model, currently on version 1.3, that Meta built for agentic work and reasoning. Muse is the consumer app and agent product that runs on top of Muse Spark to actually carry out tasks across a user's connected apps and accounts.
How much does Meta Muse cost?
Meta describes Muse as free for most everyday use, with paid plans for people who want more capacity. Independent reporting since launch has described a free tier with a weekly token allowance, a Power tier at $20 a month, and a Maximum tier at $100 a month, though Meta's own announcement did not publish exact prices.
Is Meta Muse safe to use?
Meta states Muse runs in an isolated Muse Secure VM with credential encryption, an internal approval system called Sentinel, and full audit trails, and that it does not share conversations with ad systems. Those claims should be weighed against Meta's history of privacy settlements, including a $5 billion FTC penalty in 2019 and the Cambridge Analytica scandal, rather than accepted without scrutiny.
Can Muse access my passwords or payment information?
Meta says Muse does not have direct access to user passwords and that access to each connected app is granted individually, with users choosing the level of permission, such as read-only versus send access for email. Payment actions go through connected payment systems rather than Muse holding payment credentials itself, according to Meta's own description.
Need help with this?
Growthtrait can help you put this into practice. Let's talk about your goals.
Contact us
