Making phone calls could soon become another task Americans hand over to artificial intelligence, but giving a chatbot permission to speak on someone’s behalf raises serious questions about privacy, accuracy, and what happens when technology gets something wrong. On November 12, Meta officially expanded its Muse AI phone-calling feature to all paying subscribers in the United States, allowing the assistant to contact businesses, speak with representatives, and handle everyday requests without users having to pick up the phone themselves.
The rollout follows a limited testing period that began in September, when only selected users could access the feature. Now, subscribers can ask Muse to contact restaurants, insurance providers, cable companies, and other businesses.
Meta’s chief AI officer, Alexandr Wang, announced the expansion on X, encouraging customers to “let Muse take care of all those calls you don’t want to make.”
🚨MUSE FEATURE ALERT!!!🚨
lots of you have been asking for this one…
muse can make phone calls for you! ☎️📞📲
let muse take care of all those calls you don’t want to make. insurance? cable? DMV? dinner res? not your problem anymore.
Wear the cultureShop Baller Merch →Made to order and shipped to your door.
out now to paid subs in the US 🇺🇸 https://t.co/dsyMNmI2Ub
— Alexandr Wang (@alexandr_wang) November 12, 2026
Wang suggested the assistant could tackle everything from dinner reservations to dealing with the Department of Motor Vehicles.
However, the convenience comes with a price tag. Muse subscriptions start at $20 monthly, providing 500 million AI tokens per week. A $100 monthly option includes 3 billion weekly tokens for customers who use the technology more frequently.
Meta reported in October that Muse had surpassed 10 million users, although the company has not disclosed how many are paying subscribers.
While the technology promises to eliminate frustrating customer service experiences, its expansion introduces challenges that go beyond simply getting a business to answer.
One major concern involves AI making mistakes during conversations that could have financial consequences.
For example, someone might instruct Muse to negotiate a lower internet bill, only for the assistant to misunderstand an offer, agree to different service terms, or accept additional charges without the customer realizing what happened.
A restaurant reservation could involve the wrong date, an appointment could be scheduled incorrectly, or important insurance information could be misunderstood.
These are potential risks, not confirmed failures of Muse’s newly expanded calling feature. However, they highlight why consumers may want to review the outcome of important conversations before relying on the assistant’s decisions.
Privacy presents another challenge, particularly when calls require account numbers, personal addresses, financial details, or medical information.
Those concerns are not entirely hypothetical.
In September, Reuters reported that Meta had experimented with a feature allowing human contractors to secretly handle certain calls initiated through Muse.
The arrangement reportedly sparked internal objections because employees feared sensitive customer information could reach people users never expected to access.
One employee warned that the system was “one bug away from unnecessary information being leaked to human callers.”
Meta subsequently rolled back that particular experiment, acknowledging problems with how human involvement had been disclosed. The company said appropriate safeguards would be required before bringing such capabilities to a wider audience.
It remains unclear whether human contractors have any role in the newly expanded version.
Other reported experiences have demonstrated how easily AI assistants can take unexpected actions.
The Associated Press recently documented an incident involving a Facebook Marketplace seller whose Muse assistant agreed to a transaction and shared his home address with a buyer without the seller realizing those actions had occurred.
Meta attributed the address disclosure to permissions the user had granted and acknowledged a separate pricing mistake.
Although that incident did not involve phone calls, it illustrates the potential consequences when AI assistants operate with broad authorization.
Businesses also have reasons to be cautious.
During earlier testing, some companies reportedly disconnected calls after discovering they were speaking with AI rather than an actual customer.
That reaction could become increasingly common as automated voices grow more realistic and businesses attempt to distinguish legitimate customer requests from spam, scams, and impersonation attempts.
There are also legal considerations surrounding automated communications and call recordings. The Federal Communications Commission has confirmed that certain AI-generated voice calls fall under federal telephone consumer protection rules, although the requirements depend on the circumstances.
Companies deploying calling assistants must also consider applicable privacy and recording-consent laws.
Another unresolved question is how businesses will verify that an AI assistant has permission to make account changes or discuss confidential information for a specific customer.
A convincing automated voice alone cannot establish someone’s identity or authorization.
Meta is not alone in pursuing automated phone conversations. Google’s Gemini and AI startup Instinct have developed similar capabilities, signaling growing competition to replace routine human interactions with digital assistants.
For consumers exhausted by endless phone menus, long hold times, and repeated transfers, the appeal is understandable.
Still, an AI assistant that saves 30 minutes on hold could create a much bigger headache if it shares private information, misunderstands instructions, or makes an agreement its user never intended.
