Meta has opened Muse’s business-calling feature to paid subscribers in the United States, expanding access beyond a small group of early testers. Meta chief AI officer Alexandr Wang announced the change on November 12, 2026, according to TechCrunch’s report.
The intended jobs are familiar: calling a cable company, contacting an insurer, dealing with the DMV or requesting a restaurant reservation. The appeal is letting an agent handle the phone call, including the tedious path toward someone who can help.
The more consequential detail sits in Meta’s own documentation. Its Help Center offers businesses a way to opt out of Muse calls, but gives little detail about how recipients recognize an AI caller or how that opt-out takes effect. Paid availability is now broader; successful conversations still depend on the business answering.
What Changed for Paid US Subscribers
Muse’s calling feature entered testing in September, when selected early adopters could request access through the agent. The November announcement opens it to paid US subscribers without that earlier selection step, TechCrunch reports.
That makes this an expansion of an existing feature, not Muse’s first demonstration of phone calling. Free accounts and people outside the United States are outside the announced rollout.
TechCrunch reported the following subscription options:
| Monthly price | Reported Muse token allocation |
|---|---|
| $20 | 500 million tokens per week |
| $100 | 3 billion tokens |
These are reported subscription allocations, not published call-minute allowances. The coverage does not translate tokens into a number of calls, time spent on hold or completed tasks. Buyers therefore cannot use those figures alone to calculate what each successful reservation or customer-service request will cost.
I’d treat that distinction as important before subscribing specifically for calling. Access to a feature is one thing; knowing how much useful calling a plan supports is another. The available reporting provides the entry price, but not enough call-specific usage information for a meaningful cost comparison.
The reported use cases also vary considerably in difficulty. Asking whether a restaurant has space for four people is a narrower request than resolving an insurance problem. Putting both on a list of supported intentions does not mean the agent can complete them with equal reliability.
Delegating a Call Is Different From Completing a Task
Meta presents Muse calling as a way to hand over errands that still require a phone conversation. TechCrunch describes phone trees, hold times and transfers as the friction an agent could absorb on a user’s behalf.
The practical workflow has several separate hurdles:
- Reach the right business. Meta’s Help Center describes calls to businesses with publicly available phone numbers.
- Get to the right person or service. A phone menu or transfer may stand between answering the call and addressing the request.
- Explain what the user wants. The recipient needs enough information to act.
- Meet the business’s requirements. A request involving an account may require authentication or authorization.
- Obtain a clear outcome. A connected call can still end without a booking, answer or resolved issue.
That breakdown is an analysis of the task, not a description of verified Muse performance at each stage. The opened sources provide no independent success-rate measurements, detailed operating limits or documented procedure for handing a difficult conversation back to the subscriber.
For users, the useful implication is to define the request narrowly. A restaurant booking request is clearer when it includes the date, party size and acceptable times. An account-related request needs boundaries around what the agent may agree to. Those are sensible delegation practices, not confirmed Muse setup fields.
A task that requires someone to choose between new charges, contract changes or disclosure of sensitive information carries different consequences from checking opening hours. I’d start with requests where a misunderstanding is easy to spot and reverse. That recommendation follows from the uncertainty around outcomes, not from hands-on testing of Muse.
Meta Gives Businesses an Opt-Out Route
Meta’s Help Center notice for people who do not use Muse explicitly addresses AI phone calls. It describes calls to “businesses whose phone numbers are publicly available,” with examples including checking opening hours and making bookings.
The same section links to a form businesses can use if they prefer not to receive these calls.
That is a meaningful control. A business does not need to be a Muse customer to be affected by the product, and Meta acknowledges that recipients may want to decline this form of contact.
The documentation also contains a wording mismatch worth noting. While Wang’s announcement describes paid US availability, the Help Center still characterizes calling as something Meta is experimenting with. The announcement is evidence of expanded access; the Help Center describes the recipient-facing policy. Its experimental wording should not be read as proof that calling remains restricted to the earlier test group.
The page leaves several operational details unexplained:
- How Meta verifies that an opt-out request comes from the business.
- How quickly a request becomes effective.
- Whether a business with several telephone numbers must submit each one.
- How a business can report continued calls after opting out.
Those are gaps in the reviewed documentation, not evidence that the mechanism fails. The form itself was not among the opened sources, so its fields and submission process cannot be assessed here.
The practical distinction is between having an opt-out route and knowing how that route works. Meta documents the first. Businesses still need the second to make it a dependable operational control.
Opting Out and Identifying a Caller Are Separate Problems
A business-level opt-out can reduce future contact. It cannot, by itself, explain the call already ringing at reception.
Recipients need to distinguish at least three things: who or what is speaking, which customer the caller represents, and what that customer has authorized. An AI identifying itself addresses only the first.
For example, a receptionist might be willing to discuss opening hours with an automated caller while requiring additional verification before changing a customer’s booking. An insurer could face a different authorization problem from a restaurant. These are practical examples of why the scope of a request matters, not reported incidents involving Muse.
The reviewed Meta notice does not specify the introduction Muse uses, when AI disclosure occurs or how the agent represents its authority to act for the subscriber. It also does not explain whether the recipient can request an immediate handoff to the person behind the request.
My view is that clear identification would be more useful than making the voice difficult to distinguish from a person. A realistic voice may help conversation flow, but it tells the recipient nothing about permission to change an account or accept a charge. Identity, authorization and conversational quality solve different problems.
Business Acceptance Remains a Practical Constraint
Frequently Asked Questions
4 questions
1Who can use Muse’s business-calling feature?
The November 12 announcement covers paid Muse subscribers in the United States. Previously, access was limited to selected early adopters who requested it.
