Voice cloning is the most ethically charged feature in AI agent products today. The technology is mature: 30 seconds of clean audio is enough to produce a believable clone in most cases; 10 minutes produces a clone that beats most casual listeners' detection ability. The upside for legitimate use cases is real (a small-business owner gets a 24/7 receptionist that sounds like them, without the cost of hiring). The downside for misuse is real too (deepfake fraud is now a measurable category of crime).
This post is what we think about voice cloning for business voice receptionists, how the consent and disclosure rules work, and the technical mechanics that make a difference between honest use and deceptive use. We try to be honest about what we are doing and why.
What we mean by voice cloning here
The narrow definition we operate under: a small-business owner records 30 seconds to 10 minutes of their own voice, the agent produces a synthesized version of that voice that can read text aloud, and the synthesized voice is used as the business's voice receptionist on inbound calls.
What we are NOT doing: cloning someone else's voice without consent. Cloning a celebrity's voice. Cloning a deceased person's voice. Producing audio of someone saying things they did not say. These uses exist in the broader voice-clone-tech category; they are not what Ajeris ships and they are explicitly forbidden by our terms.
What is legal in 2026
US law on voice cloning is jurisdiction-specific and changing. The general framework as of 2026:
Federal level. No comprehensive law. The FCC has clarified that AI-generated voices used in robocalls require the same disclosures as human-made robocalls. The FTC has authority to act against deceptive uses under unfair-and-deceptive-practices doctrine.
State level. Several states (Tennessee, California, Illinois, others) have enacted right-of-publicity laws specifically covering AI voice clones. Most of these protect the right to license your own voice; they do not prevent you from cloning your own voice for legitimate use.
Practical bar. Cloning your own voice for your own business is legal in every US jurisdiction we are aware of, with consent flowing only one way (you to yourself). Cloning someone else's voice requires their explicit, written consent. Disclosure to callers is required in some jurisdictions; we treat it as good practice everywhere.
What we require
Three things, when you set up voice cloning in Ajeris.
- 1Consent statementYou record a specific phrase confirming you are cloning your own voice.
- 2Identity verificationThe owner of the cloned voice has an account; the clone is bound to that account.
- 3Use case attestationYou attest the use is your own business; misuse violates terms and removes the clone.
The consent statement is the part most products skip. We require a recorded phrase ("I am authorizing Ajeris to use my voice for my business AI receptionist") not because it is legally required everywhere but because it creates a clear evidentiary record if the legitimacy is ever challenged.
The disclosure question
A real ethical question that a binary law does not resolve: should the agent disclose to callers that it is AI?
Three views in the industry.
The "mandatory disclosure" view. Disclose every call, in the greeting. "Hi, this is Maya, AI assistant for Sterling Cuts." This is the most ethically conservative position; it also reduces the value because some callers immediately disengage upon hearing "AI."
The "no disclosure unless asked" view. The agent identifies as the business name in the greeting; it discloses if directly asked ("am I talking to a real person?"). This is what most voice-receptionist products in 2026 default to.
The "deceptive disclosure" view. The agent claims to be a human even when asked. This is what bad actors do; we explicitly do not do this.
Ajeris defaults to the middle position. The agent identifies as the business name, in the cloned voice. It does not volunteer that it is AI. It does respond honestly when directly asked. The owner can change the default to mandatory disclosure if their business posture requires it.
What about the caller's right to know
A reasonable position: the caller has a right to know they are talking to an AI. The counter-position: callers are talking to "the business," not specifically to a human, in the same way a chatbot interaction is not commonly disclosed as "this is software." The cultural expectation is shifting; in five years the default may be different.
For now, we recommend businesses match disclosure to context. A high-trust legal practice may want mandatory disclosure. A salon's appointment booking probably does not. The decision is the owner's.
How the cloning technology actually works
A simplified version. The agent takes your voice sample and learns the acoustic characteristics: timbre, pitch range, pace, common-phoneme prosody. At inference time, when the agent has text to speak, it generates audio that matches your acoustic profile.
The quality scales with the input.
- 30-second sample: instant clone, good enough for most operational calls. Some artifacts on uncommon phonemes.
- 10-minute sample (PVC, professional voice clone): studio-grade output. Indistinguishable from your real voice in most casual listening.
The voice synthesis runs on industry-standard infrastructure. The voice model is trained on your sample and stored bound to your account. Deletion is a single command and removes the trained model from our systems and the underlying provider within the documented retention window.
What we explicitly forbid
A short list, from our terms.
- Cloning anyone other than yourself or a person who has given explicit written consent.
- Using the clone to claim to be a human when the caller has directly asked if they are talking to a human.
- Using the clone for any deceptive financial purpose (impersonating a CFO to authorize wires, etc.).
- Using the clone to harass or threaten anyone.
- Using the clone in any context the cloned person has not consented to (a former employee whose voice was once cloned for the business cannot have that clone re-used after they leave).
Violations can result in immediate account termination and, in serious cases, referral to relevant authorities.
- Owner clones their own voice with explicit consent.
- Agent identifies as the business in greeting.
- Agent answers honestly if asked 'am I talking to AI?'
- Use is operational (booking, invoicing, customer care).
- Cloning someone else's voice without consent.
- Misrepresenting the AI as a specific named human.
- Lying to callers about whether they are talking to AI.
- Use for deceptive financial purposes.
Frequently asked
Can I clone an employee's voice for the business?
What if a customer is upset that they were talking to AI?
Is the clone deepfake-detectable?
What about using a stock voice instead?
What about regulatory disclosure rules in California / Illinois?
Try it
The honest test is to clone your own voice and call yourself. Listen to the recording. Decide for yourself whether you would be comfortable using this for your business. The answer is usually yes; the discipline is in the use, not the technology.
