Claude Duet is "always on", but it checks your accounts every few hours: what Anthropic, OpenAI and Meta's background agents will do without asking

Anthropic opened a waitlist on 21 October for Claude Duet, an app that reads your connected email, calendar and business tools and messages you when something needs attention. Its product page calls it "always on"; its launch announcement says it checks your accounts every few hours, because checking nonstop would use up a subscriber's plan limits much faster. It follows OpenAI's Dots and Meta's Muse, and the three draw the line between acting and asking in different places.

Is Claude Duet always on?

It is always available, and it keeps working on tasks you give it, but it does not watch your accounts continuously. Anthropic's product page says: "While it's always on, it uses good judgment when it reaches out and knows what can wait." The launch announcement is more specific: "Claude Duet checks your connected accounts every few hours, because checking nonstop would use up your plan's limits much faster." You can tell it what you want to hear about and how often.

So "always on" describes the agent, not the monitoring. If an urgent email lands just after a check, Duet may not see it for hours. Anthropic deserves credit for saying so: OpenAI describes its dots as "always-on" and "available 24/7," and Meta says Muse "keeps working after people close the app," but neither publishes how often their agents look at your apps unprompted.

What will Claude Duet, ChatGPT Dots and Meta Muse do without asking?

Read, sort and prepare. All three run on their own cloud computer with a browser, so your laptop or phone does not need to be on.

• Claude Duet "reads, drafts, and prepares work on its own," Anthropic says. It sends a morning brief on the most important items across your connected apps and messages you during the day "only when something needs your attention." It carries over your memory, skills and connected apps from your Claude account.
• ChatGPT Dots run what OpenAI calls proactive research when you are not working with them. Per OpenAI's help center, it "can read information from permitted connected sources and save private notes, but its research tools cannot directly" send messages to other people, change content through plugins, or control a browser or computer.
• Meta Muse "can make suggestions unprompted and act on details that person only mentioned once," Meta says. Its example: turning a recipe reel saved on Instagram into a grocery list and remembering friends' dietary restrictions before sending dinner invitations.

Dots are the most explicit about the limits of unprompted work: the read-only restriction on proactive research is built in, and OpenAI says Custom Rules cannot change it.

What do Duet, Dots and Muse wait for your approval on?

Anything that sends, pays or deletes, with different wording in each case.

Duet "shows you exactly what it's about to do and waits for your approval," Anthropic says. Dots start with default rules that you can tighten or loosen with Custom Rules, and you can approve some actions in advance, such as recurring messages; "Approving one message does not give your dot ongoing permission to contact people on your behalf," OpenAI notes. Muse routes everything through a separate Sentinel agent: "Nothing Muse does reaches the internet unless the Sentinel approves it, and it asks the person for permission when needed."

What will they never do, even with your approval?

Each company draws one hard line, and they are not the same line.

• Claude Duet: "A few things are off-limits even with your approval, like trading securities or moving money between accounts." Its new sensitive-information classifier is designed to reject requests to share the strictest tier, "including credit card numbers and government identification." Health or personal finance details go out only when you have explicitly told Duet to send them.
• ChatGPT Dots: "The most sensitive actions like changing a password or transferring money require you to take over so you can complete them yourself." Your approval "cannot override core safety requirements."
• Meta Muse: Meta does not publish a list of actions refused even with approval. What it does publish is a filter: Muse's email connector "filters out one-time tokens, password reset links, and login magic links," so connecting your inbox does not let the agent, or someone steering it, sign in as you elsewhere.

All three say the agent never sees your passwords. Duet's credential handling follows Anthropic's Managed Agents design; Dots pause for you to type into a secure login form; Muse stores credentials outside the agent and inserts them at the network boundary.

Who can use Claude Duet, ChatGPT Dots and Meta Muse?

Only Dots are available outside the US, and in the EU only on the Business Premium plan.

• Claude Duet: a research preview by waitlist, starting with Pro and Max subscribers in the US, on iOS, Android and claude.ai, in its own app separate from the main Claude app.
• ChatGPT Dots: rolling out "to Pro users in markets excluding the European Economic Area, Switzerland, and the UK," and to "Business Premium users across all supported ChatGPT regions." Enterprise, Edu and Healthcare get a beta that admins must switch on. Not available to users under 18.
• Meta Muse: "rolling out in the US on iOS, Android, and muse.ai," and reachable in WhatsApp. Meta says it is "free for most of what people need, with subscription plans for people who want to do more."

Do Claude Duet's background checks use up my plan limits?

Yes. Duet "shares your plan's usage limits with the rest of Claude," Anthropic says, and "Its background checks count toward those limits." Larger tasks, like a financial model or a deck, take more model turns. That is the reason it checks every few hours rather than constantly.

OpenAI prices it differently. Your first dot is included in Pro or Business Premium at no extra cost, its chats with you do not count toward your ChatGPT usage limits, and the plan includes a separate allowance for deeper work, with higher limits in the first month. Tasks a dot starts in Codex or ChatGPT Work count as usual.

Are Duet, Dots and Muse conversations used to train AI models?

It depends on your settings, and only OpenAI addresses the background work directly.

• ChatGPT Dots: on personal plans, the "Improve the model for everyone" setting decides whether dots' conversations and work can be used. "We don't train directly on proactive background research or its notes," OpenAI says, though information a dot brings into an eligible conversation may be used. Business, Enterprise and Edu data is not used for training by default.
• Meta Muse: people "can also opt out of their interactions being used to train Meta's AI models," and Muse does not share conversations or VM data with Meta's ad systems.
• Claude Duet: the announcement is silent on training. Pro and Max are consumer plans, where Anthropic's Privacy Center says it uses chats for training if you choose to allow it, with exceptions such as conversations flagged for safety review. It has not said how that setting applies to Duet's background checks.

What can go wrong with a background AI agent?

The companies list the risks themselves. Anthropic says Duet "still might misread a thread or miscalibrate what's urgent," is weaker on personal errands such as booking tickets, and "can't yet show you its browser." OpenAI says dots "can still make mistakes, so always review consequential work," and that its prompt-injection protections "help reduce the risk" but "do not eliminate it." Meta's safety post says it "designed the system to assume the agent may be under attack and limit the potential damage."

The practical difference for a reader is the one in the headline. An agent that checks every few hours is a good assistant for a morning brief and a weekly backlog, and a poor alarm for anything that cannot wait. Anthropic is the only one of the three to say how often its agent looks.