Can Muse Make Phone Calls? Now It Can With Its Own Number
Can Muse make phone calls for your ops workflows, learn its real limits before you build around it and waste cycles.
Yes, Muse can make phone calls. But access is limited, the scope is narrow, and the gap between personal errand tool and business calling infrastructure is bigger than most pilots account for.
Muse's phone-calling capability is real, and the answer to the title question is yes. But "yes" carries a significant qualifier: as of the incremental rollout that nullbot newsroom reported in October 2026, the feature launched in limited early access around September 2026 and is not available to every account. The common assumption is that any AI agent that can place a call is architecturally ready for business use, that the calling capability IS the product. The gap between that assumption and the reality is worth understanding before you build any workflow around it.
For operations and CX leaders, that gap matters more than it might seem. A consumer AI agent that dials one call at a time for personal errands sits in a different category from the dedicated calling infrastructure a team needs for high-volume, auditable business calls. Conflating the two causes premature pilots and wasted cycles.

Muse's phone-calling feature exists, but most users asking about it today will find it unavailable. According to nullbot newsroom's October 2026 coverage, the rollout began around September 2026 on an incremental basis, with users who had already requested early access prioritized first. That structure is deliberate: Meta is stress-testing the feature in controlled conditions before widening availability.
The scope is also narrow by design. Meta's Muse calling feature is scoped to calls placed to U.S. businesses on the user's behalf. International businesses, personal contacts, and anything outside that defined boundary fall outside what the feature currently handles. If you requested access and didn't receive it, that is the expected experience for most users right now, not a technical error.
According to nullbot newsroom, the interaction types the feature targets are precisely: booking a restaurant, checking business hours, moving an appointment, or confirming a delivery window. Low-stakes. Scripted. Predictable. The feature works best when the call follows a recognizable pattern and the outcome is simple to verify.
The practical ceiling becomes clear fast. Muse handles the occasional personal errand call. It is built for individual users, not for the thousands of concurrent, regulated, auditable voice interactions a CX or ops team runs daily. That is an architectural distinction, and it is worth understanding exactly which calls it can handle before deciding whether it fits your workflow.
Key takeaways#
- Muse can make phone calls, but only a tightly scoped set: restaurant reservations, business hours checks, appointment rescheduling, and delivery confirmations. That list is the product, not a starting point.
- The feature launched in limited early access around September 2026 and is still not available to every account, which means most users asking 'can Muse make phone calls' are asking about something they cannot yet access.
- A human concierge layer sits inside Muse's calling flow, the AI does not operate autonomously end-to-end, and that distinction matters the moment calls carry commercial or compliance stakes.
- Privacy concerns raised by Muse's calling feature are not a consumer story. They expose a structural gap between an AI that can dial and a platform built to handle what happens when that call touches regulated data.
- Concurrency is the number that separates a demo from a deployable system. A consumer agent dialing one call at a time is a different category from voice infrastructure, not a smaller version of it.
- Bland closes that gap: purpose-built AI phone calling infrastructure with fully self-hosted voice agents, dedicated numbers, and the compliance architecture enterprises actually need to scale beyond a pilot.
What Types of Calls Muse Can Actually Make on Your Behalf#
Muse can handle a tightly defined set of transactional calls: restaurant reservations, business hours checks, appointment rescheduling, and delivery confirmations. That list is the product. IVR systems have long handled exactly these categories of calls, appointment scheduling, store hours, and service requests, because they are predictable, script-friendly, and low-stakes, and businesses that deploy them well see measurable lifts in customer experience without adding headcount. The moment a call requires judgment, negotiation, or any guaranteed outcome, Muse declines to proceed.
Understanding how Muse makes calls matters as much as knowing what it dials. When you ask Muse to book a table, it places an outbound call, listens for a human or automated greeting, and works through whatever phone tree it encounters. IVR phone trees route callers based on spoken or keypad input, and Muse is designed to parse those branches and select the right path. It will also wait on hold, which is the feature most users notice first, because hold queues are where personal call tasks die. The practical value is real: delegating a call you would otherwise abandon after seven minutes of hold music is a genuine time recovery.
The refusal boundaries are equally real and worth naming plainly. Muse will not make emergency calls, medical advisory calls, or any call where the outcome needs to be guaranteed. Automated systems are explicitly not suited for emergencies or complex advisory interactions that require human judgment, and that constraint applies directly to AI calling agents acting on a user's behalf. If the call could go wrong in a way that harms someone, Muse steps back.
Access is also gated. The feature remains in limited beta, with earlier requesters prioritized. Most accounts cannot use it yet.
The deeper issue sits beneath the feature itself. Muse's calling capability was never fully autonomous. Meta embedded human concierge workers to complete calls the AI alone could not finish, partly because businesses were hanging up on AI voices. That human fallback was not disclosed to users, and it disappears entirely if you try to run more than one call at a time. For an individual delegating a single reservation, that is a footnote. For any operation that needs to run calls at volume, it is a structural ceiling.
When an operations team needs outbound calling at scale, the question is whether the platform behind it can handle hundreds of concurrent calls, maintain an audit trail, and keep data from flowing through third-party providers mid-conversation. Those are different problems entirely, and consumer AI calling tools are not built to solve them.
Platforms designed for high-volume operations, where the goal is to handle calls 24/7 without adding headcount and automate high-volume, high-stakes calls end to end, are built on fundamentally different architecture: concurrency sized to actual demand, per-minute pricing that absorbs transcription and voice costs, and integration layers that sit on top of existing contact center infrastructure like Amazon Connect without requiring a platform migration. Bland, for instance, is most beneficial when a business already has a contact center platform or CRM and wants to layer AI calling on top without replacing existing infrastructure, a meaningful distinction from consumer delegation tools that operate as standalone experiences.
But cataloguing what Muse can dial is only half the picture. The more revealing question, one that changes how you should evaluate any AI calling tool, is who, or what, is actually on the line when Muse picks up the phone for you.
1. Phone Tree Navigation, Hold Waiting, and Callback Handling#

Muse is designed to work through the friction that makes most people abandon calls before they connect. When it reaches an automated system, it parses spoken prompts and keypad menus, selecting the correct branch at each step. IVR systems route callers based on spoken or keypad input, and Muse is built to traverse that same infrastructure.
It will also wait on hold without dropping. That is the practical win most users notice first: a call you would have abandoned after eight minutes of hold music gets completed without your attention. The limitation worth naming is that Muse handles one call at a time.
If the hold queue runs long and a second task needs a call placed simultaneously, it waits.
Operations teams that regularly lose hours to hold queues across dozens of simultaneous campaigns face a categorically different problem. Businesses in that position, running outbound sales calls, follow-up sequences, or inbound support coverage around the clock, need concurrency measured in the hundreds, not a single-threaded consumer agent. A platform built for 100 concurrent calls with a 99.9% uptime SLA and a daily cap of 5,000 calls is solving a different equation than a consumer tool waiting in a single queue.
2. Reservations, Cancellations, Hours Checks, Rescheduling, and Delivery Confirmations#

Muse targets calls that follow a predictable script with a clear endpoint. Restaurant reservations, appointment moves, subscription cancellations, store hours checks, and delivery window confirmations all fit that pattern. IVR systems have long handled exactly these categories because they are low-stakes and repeatable. The tradeoff is specificity: Muse performs well when the call has a defined success state ("reservation confirmed for two at 7pm") and struggles when the outcome depends on negotiation, pricing discretion, or a representative making an exception. If your call requires the person on the other end to bend a policy, Muse is not the right tool.
For businesses running the same call types at volume, appointment confirmations across a multi-location practice, delivery window checks across thousands of orders, the individual-task framing of a consumer tool stops being useful. High-volume, high-stakes calls that run continuously benefit from conversational pathways, knowledge bases, and integrations that connect directly into existing CRM or contact center infrastructure. The calls are the same category. The infrastructure requirements are not.
3. Limited Beta Access - How to Check Eligibility and Request It#
The calling feature is not available to all Muse users. Meta has rolled it out incrementally, prioritizing accounts that requested early access first. Most users checking today will find the option absent from their interface entirely.
Call capability alone is insufficient without the contextual data access and back-end integration needed to complete tasks meaningfully, which partly explains the cautious rollout pace. The practical guidance: check your Muse settings for a calling or phone option, and if it is absent, the waitlist is the only path. There is no public timeline for general availability, so planning any workflow around this feature today carries real scheduling risk.
4. Calls Muse Will Refuse - Emergencies, Medical Advice, and Guaranteed Bookings#

Muse has explicit refusal boundaries, and they are not soft suggestions. It will not place emergency calls, provide medical advisory guidance over the phone, or make any call where the outcome must be guaranteed. Automated systems are explicitly not suited for emergencies or complex advisory interactions requiring human judgment, and that constraint applies directly to AI calling agents acting on a user's behalf.
The refusals are sensible design. The problem surfaces when users assume the boundary is narrower than it is. Any call with legal, medical, or safety stakes falls outside scope, and any call where "it didn't go through" has real consequences for the person waiting on the other end should not be delegated here.
One pattern operations teams frequently run into is the assumption that a capable consumer AI calling tool will scale sideways into business use cases without friction. It does not. The calls that matter most in a business context, outbound appointment reminders that directly affect no-show rates, inbound intake calls for regulated services, follow-up sequences tied to revenue, are exactly the calls where guaranteed availability, audit trails, compliance documentation, and integration with existing systems are non-negotiable. Those requirements describe a different class of platform entirely.
How the Human Concierge Layer Inside Muse Calls Works#
62%
of business calls go unanswered

Human Contractors Were the Voice Behind Some Muse Calls#
The Wall Street Journal reported in May 2025 that Meta used this hybrid model during early Muse calling tests, and that the arrangement was not disclosed to users at launch.
Why Businesses Hanging Up on AI Voices Forced Meta's Hand#
The Wall Street Journal's May 2025 reporting noted that businesses were frequently hanging up when they detected an AI voice, and that the human concierge layer was partly designed to work around that rejection pattern. A human voice could clear that bar. An AI voice, at least in 2024 and early 2025, often did not.
Key takeaway: The answer to call rejection is a voice pipeline built to complete complex, multi-step calls end-to-end, without the opacity.
Bland AI's approach is designed around exactly that requirement: automating highest-volume, highest-stakes calls end-to-end, with real-time transcription and premium voice quality included in the per-minute rate across every plan, from the Start plan at $0.14/min up through the Scale plan at $0.11/min, so the economics of full automation are legible from day one, with no undisclosed labor cost hiding behind the product.
The Undisclosed Handoff and the Pullback That Followed#
Meta did not tell users that a contractor might be on their call. According to The Wall Street Journal's May 2025 coverage, that omission raised internal concerns among Meta employees, which triggered a re-evaluation of the feature, and the feature was subsequently reported as withdrawn or suspended.
Bland's Enterprise plan addresses this directly: compliance documentation is available under NDA, data residency controls are available, BAA is available, and a forward-deployed engineering team operates under a 28-day deployment framework that scopes, builds, and gray/red/green-team tests every agent before go-live. That transparency extends to how AI calling fits into a business's existing infrastructure.
For teams already operating inside Amazon Connect or a broader CCaaS stack, that question has a clear operational answer when AI calling is integrated directly with existing infrastructure. Bland's integrations platform connects AI calling into the CRM, CCaaS, and telephony systems a business already runs, so the call record, the transcript, and the data flow all live exactly where compliance and operations teams expect them to.
What Privacy Concerns the Muse Calling Feature Actually Raised#
The concerns raised about Muse's calling feature were not primarily about AI placing calls at all. They were about what was happening inside the pipeline that placed them, and whether the businesses and users on either end of those calls had any visibility into it.

The Undisclosed Human Layer Was the Core Privacy Breach#
Capability and compliance are distinct thresholds, and Muse's feature appeared to satisfy only the former.
Every Word Spoken on a Muse Call Passed Through a Frontier Model's Infrastructure#
Routing call audio through a third-party frontier model provider creates an exposure surface most compliance teams cannot fully audit. State-level AI disclosure laws in California, Utah, and Colorado impose affirmative obligations on businesses deploying AI voice agents, meaning the compliance burden is materially heavier for a business running thousands of calls than for an individual consumer booking dinner.
Consent, Liability, and the Unanswered Question of Who Owns a Mistake#
Businesses deploying AI calls must address consent, disclosure timing, liability, and data handling before launch, because the calling capability alone does not satisfy legal readiness. For enterprises operating at scale, those gaps compound quickly.
Bland's Scale plan supports up to 100 concurrent calls, 1,000 per hour, and 5,000 per day, with 100 knowledge bases and version-locked agents so the behavior your compliance team approved is the behavior that runs.
How Muse Compares to Other AI Agents Making Phone Calls Right Now#
Muse vs. Other AI Phone Call Agents#
Calling capability is table stakes. The real question is what sits underneath it, and that is where consumer AI agents and purpose-built voice platforms stop being the same category.
"I'm concerned that Muse makes phone calls on behalf of users without identifying itself as an AI, raising legal and ethical questions compared to other AI phone call agents that do disclose their AI nature."
— what we hear from AI assistant users
When Muse and Instinct launched phone calling features at roughly the same time, the framing was competitive: two consumer AI agents racing to add a headline capability. That framing is accurate as far as it goes. But it obscures the dimension that actually matters for business deployment. As caller.digital's voice AI benchmarking makes clear, the ability to place a call is only the baseline; businesses evaluating AI calling agents must assess concurrency limits, audit trail completeness, data routing controls, SLA guarantees, and compliance documentation, none of which consumer agents typically publish.
The concurrency gap alone is disqualifying for most operations teams. Purpose-built voice AI platforms support hundreds to thousands of simultaneous calls by design, whereas consumer agents like Muse impose undisclosed or low concurrency ceilings. A team that discovers this mid-pilot has wasted weeks and credibility.
The compliance gap is the sharper risk. FCC rules, state-level AI disclosure laws, and TCPA consent chains impose obligations on businesses deploying AI calls that do not apply to individuals using a consumer app. That distinction is a structural barrier, and legal readiness is a precondition before a single production call is placed.
Use this five-dimension framework to stop evaluating these tools as if they belong in the same category.
1. Bland - Best Enterprise Voice AI for High-Stakes, Self-Hosted Calling#
Most operations teams reach this point and default to the most recognizable consumer agent, reasoning that a pilot will surface the gaps. In practice, the gaps surface after the pilot has consumed budget and credibility. Bland's fully self-hosted architecture keeps every call, transcript, and PII record inside infrastructure the business controls, with documented SLAs and compliance posture that consumer agents cannot match. This is most relevant when your operation runs high call volumes in a regulated industry and your legal team needs a compliance package it can actually sign off on. The honest trade-off: the infrastructure depth that makes Bland appropriate for enterprise scale is overkill for a single user booking a restaurant reservation.
2. Meta's Muse vs. Instinct - Two Consumer AI Agents Racing to Own Phone Calls#
Muse and Instinct launched phone-calling capabilities within the same news cycle, signaling a competitive land-grab among consumer AI agents rather than a deliberate enterprise rollout. Both can handle low-stakes tasks like restaurant reservations or subscription cancellations. Neither was architected for regulated industries, audit trails, data routing controls, and compliance documentation are absent, making them unsuitable for business-critical call workflows.
3. Concurrency, Audit Trail, and Data Routing - The Evaluation Dimensions That Actually Separate Voice AI Platforms#
Asking whether an AI agent can dial a number is the wrong question for enterprise buyers. The dimensions that matter are: concurrent call capacity under load, immutable audit trails for regulatory defensibility, configurable data routing to control where audio and transcripts land, formal SLA commitments, and compliance documentation (SOC 2, HIPAA, GDPR). Consumer agents like Muse score poorly on all five; purpose-built platforms are designed around them.
4. Privacy, Compliance Risk, and Ethical Exposure - The Hidden Cost of Consumer Agent Phone Calls#
When Muse or Instinct places a call on a user's behalf, the audio traverses shared infrastructure with opaque data-handling policies, a material risk under TCPA, GDPR, and state wiretapping laws. Purpose-built enterprise platforms document consent flows, call recording disclosures, and data retention policies explicitly. For any organization where a mishandled call creates legal liability, the compliance documentation gap in consumer agents is not a minor inconvenience but a disqualifying factor.
Why Enterprises Scaling AI Calls Need Infrastructure Not Just an Agent That Can Dial#
Bland AI's noise cancellation can be configured at three levels: the agent (Persona) level, per individual call, or per inbound phone number.

Our own research found that evals are positioned as a QA and compliance scoring tool for teams that need to audit failure modes across calls at scale without manual intervention.
Our data shows that most TTS models are trained on professional recordings such as audiobooks, podcasts, and voiceovers, which teach polished cadence but not the fragmented, self-correcting nature of real conversation. In our own words: "Many speech models learn from professional recordings: audiobooks, podcasts, voiceovers, narration, and carefully staged studio reads."
The Stack Beneath the Call Is Where Enterprise Deployments Break#
Key takeaway: Voice AI systems must be architected to handle peak call volume, not average load. A system sized for average demand will fail during surges.
The Enterprise AI Calling Infrastructure Checklist#
Broader industry trends show that platforms owning their own FCC-licensed audio bridging carrier enforce concurrency inside the platform rather than inheriting a third-party CPaaS ceiling, a distinction that matters enormously when a campaign needs 500 simultaneous lines.
- Concurrency provisioned at the carrier level, not as a software cap on a rented network
- 99.9% uptime SLA
- Defined data residency
- Full call transcripts stored with tamper-evident audit trails
- Compliance documentation available under NDA
- HIPAA documentation for regulated healthcare calls
- TCPA compliance coverage for governed outbound campaigns
- GDPR-scoped documentation for EU customer interactions
Why Full Voice Stack Ownership Wins Higher-Stakes Deals#
Infrastructure-first vendors are winning higher-ACV deals because they eliminate the operational risk that third-party-dependent platforms cannot, a pattern reflected across the market, where enterprises explicitly cite data-routing opacity and concurrency ceilings as pilot failure points.
Enterprise plans provision concurrency sized to actual volume with dedicated infrastructure, on-premises or VPC deployment, and compliance documentation available under NDA.
The architecture earns its cost when call volume is high, the regulatory environment is real, and the consequence of a compliance failure or an outage mid-campaign is measured in revenue or legal exposure, not inconvenience.
Next steps#
If your team has been evaluating Muse's calling feature as a path to business-grade AI telephony, the path forward starts with recognizing that calling capability and call infrastructure are different products entirely.
The undisclosed human concierge layer inside Muse calls reveals the first gap: a consumer agent that routes calls through contractors without disclosure cannot produce the audit trail a compliance team will accept. The concurrency ceiling reveals the second: Muse enforces one call at a time by design, while a production operation hitting hundreds of simultaneous contacts needs capacity provisioned at the carrier level, not borrowed from a consumer app. Together, those two constraints point to a platform built from the ground up for volume, auditability, and data routing you control.
Start with a dedicated AI phone number on a plan sized to your actual call volume. From there, you can configure inbound and outbound calling with real-time transcription, version-locked agent behavior, and per-minute pricing that bundles STT, TTS, and LLM inference so cost-per-contact is legible before a campaign launches.
Frequently Asked Questions#
Why can't I find the phone calling option in Muse?#
The feature is still in limited early access and is not available to all accounts. Meta rolled it out incrementally starting around September 2026, prioritizing users who had already requested early access, so most people checking today will simply not see the option in their settings. The only path forward is the waitlist, and there is no public timeline for general availability.
What kinds of calls will Muse actually refuse to make?#
Muse will not place emergency calls, provide medical advisory guidance over the phone, or make any call where the outcome must be guaranteed. Any call with legal, medical, or safety stakes falls outside its scope, and any call where a failed connection has real consequences for the person waiting should not be delegated to it.
Was a real human sometimes on the line when Muse made calls on my behalf?#
Yes. The Wall Street Journal reported in May 2025 that Meta used human contractors to complete calls that the AI alone could not finish, and this arrangement was not disclosed to users at launch. That meant a person could be listening to a call a user believed was handled entirely by software, which raised internal concerns at Meta and triggered a re-evaluation of the feature.
What precautions should I take before letting an AI agent make calls for me?#
The post highlights that sensitive information, such as scheduling context, intake data, or personal preferences, shared during an AI-placed call may be heard by a human contractor without your knowledge, as documented in Muse's case. Before delegating any call, it is worth confirming whether the platform discloses who or what is on the line, whether the data flow is fully owned by you, and whether the call types involved carry legal, medical, or safety stakes that put them outside what the AI can appropriately handle.
Can Muse handle calls for a business running hundreds of outbound calls a day?#
No. Muse is scoped to individual users placing one call at a time for personal errands like restaurant reservations, hours checks, appointment rescheduling, and delivery confirmations. It handles a single call at a time, lacks concurrency for high-volume use, and does not provide the audit trails, compliance documentation, or CRM integration that business-scale calling requires, those are fundamentally different infrastructure problems.