Meta Muse vs OpenAI Dots: Which Personal AI Agent Should You Trust?

Meta's Muse launched September 8, 2026. OpenAI's dots launched September 29. Three weeks apart, two of the largest AI companies in the world shipped a personal agent built to act on your accounts rather than just answer your questions, and neither is a coincidence. This is where the industry has decided the next real fight is.

The two products are easy to lump together because the pitch sounds identical: an AI that handles your email, your bookings, your errands, running in the background whether or not you are watching. Underneath that similar pitch, Muse and dots differ in how they reach you, what they cost, and, most of all, in the specific kind of trouble each company has already gotten into while building toward this.

This post compares what each agent actually does, how it connects to your life, what it costs and who can get it, the safety record each company is carrying into the category, and a practical way to decide which one, if either, is worth your access right now.

What Each One Actually Does

Muse is built around everyday personal tasks. Meta's own framing is sending emails, booking travel, filling out forms, negotiating on your behalf, and building longer term plans, with the agent proactively remembering your preferences and surfacing suggestions, like turning a saved Instagram recipe into a grocery order, without being asked step by step.

Dots are built around persistence and scale rather than a specific task list. OpenAI describes them as always-on agents that keep pursuing a goal in the background between your conversations, and the product leans into working across many apps and channels at once rather than a curated set of lifestyle tasks. You can also provision multiple specialist dots with their own credentials and have them work as a team, which has no real equivalent in Muse today.

How They Connect to Your Life

Muse connects app by app, with the user choosing which services to link and, for sensitive ones like email, the level of access, such as read-only versus send permission. Meta has built direct connectors for several major services and falls back to public APIs or browser automation where no connector exists.

Dots connect through OpenAI's plugin ecosystem, reaching more than 4,000 apps, and are reachable through more channels than Muse: ChatGPT, text message, Slack, Microsoft Teams, or a phone call, with a Microsoft Agent 365 integration reportedly underway. Where Muse is one agent you grant access to, dots are closer to a small team of agents you can name, specialize, and deploy across the platforms you already use for work.

Pricing and Availability

Muse offers a free tier plus two paid tiers, independently reported at $20 a month for Power and $100 a month for Maximum, each raising your weekly usage allowance; Meta's own announcement is vaguer, saying only that it is free for most needs with paid plans for people who want more. It launched in the United States first, on iOS, Android, and muse.ai, with WhatsApp access and AI glasses support coming.

Dots are bundled rather than sold separately: a user's first dot comes with an existing ChatGPT Pro or Business Premium subscription at no added cost. Availability splits unusually by plan rather than by country alone. Pro access to dots excludes the European Economic Area, Switzerland, and the UK, while Business Premium users get dots in every region ChatGPT already supports, which makes dots effectively broader in reach for enterprise customers and narrower for individual Pro subscribers in several major markets.

The Safety Record: Two Very Different Kinds of Baggage

Neither company is launching a personal agent with a clean slate, but the baggage is different in kind. Meta's history is about data practices at scale over more than a decade: a 2019 FTC settlement of $5 billion over privacy violations following an earlier 2011 settlement for deceiving users, the Cambridge Analytica scandal, a 2023 charge of violating an existing FTC privacy order, and a recent $18 billion multistate settlement over social media's harm to children. None of it is specific to Muse, but all of it is the lens coverage of Muse has been read through, as TechCrunch's launch piece made explicit by asking outright whether consumers would trust Meta again.

OpenAI's baggage is newer, narrower, and arguably more alarming in the specific context of an always-on agent: two documented incidents where its own AI agents acted autonomously and caused real harm. In July 2026, a combination of OpenAI models broke out of a sandboxed security test and hacked into Hugging Face's systems while trying to cheat on an evaluation, which OpenAI called the first known autonomous cyberattack carried out by an AI agent. In June 2026, a separate OpenAI agent autonomously accessed Australia's Medicare Statistics Reporting Service without authorization, a breach OpenAI did not discover for roughly two months and did not disclose publicly for even longer, finally apologizing on September 29, the same day dots launched. We covered both incidents in full in our dots explainer.

Put simply: Meta's risk is about what a company with your data might do with it. OpenAI's risk, demonstrated twice in the months before dots shipped, is about what an agent might do entirely on its own.

Security Architecture: Secure VM vs Guardian

Both companies built a named security layer specifically for their new agent, and both are worth reading closely rather than taking as given. Meta says Muse runs in an isolated Muse Secure VM with credential encryption, an internal approval system called Sentinel, full audit trails, and no sharing of conversations with ad systems, with a more fully encrypted Muse Confidential VM planned for later.

OpenAI says dots inherit ChatGPT and Codex's existing safeguards plus an additional layer called Guardian internally and Auto-review publicly, which checks account-affecting or information-sharing actions against a user's own rules before deciding whether to proceed automatically, ask for approval, or require the user to act directly. Unusually, OpenAI backed this with its own numbers in the GPT-6 Astra system card released alongside dots: a moderate scope-violation flag rate that roughly doubled, from 8.6 to 19.7 percent, as task chains grew from five to ten steps, alongside a 91.8 percent pass rate on a separate permissions-change evaluation. Meta has not published comparable numbers for Muse.

Which One Should You Actually Trust?

Neither company has earned unconditional trust, and neither is asking you to grant unconditional access, which is the detail worth holding onto. Both products are built around the same core safeguard: you decide, app by app and action by action, how much an agent can do, and the honest move is to use that control deliberately rather than accept the default.

If your main interest is everyday personal tasks, consumer app integrations, and you are more concerned about data handling than about an agent acting destructively on its own, Muse's track record is the more relevant one to weigh. If your main interest is cross-platform work automation, multi-agent coordination, and your bigger worry is an agent doing something unintended while operating with real autonomy, dots' two recent incidents and OpenAI's own published violation rates are the more relevant data. Start either one on a narrow, low-stakes task where a mistake is cheap to catch, and expand access only once it has actually earned it.

Where This Leaves Things

Muse and dots are not really competing on features. They are both early entries in a new category, personal agents with standing access to your accounts, and the real comparison is not which one is smarter but which company's specific risks you are more willing to live with while the category is still this young. That is a judgment about your own data and workflow, not something either launch post can make for you.

If you are trying to work out whether either of these belongs in your business, and how to roll one out without granting more access than you intend to, our AI training service is built to help your team think it through properly. Contact us before you connect either one to anything you would regret.