A voice clone is an AI-generated copy of a real person's voice that can say words they never spoke. Commercial tools need about a minute of clean audio, so any executive with public interviews can be imitated. For security teams, voice cloning turns vishing from a generic scam into a call that sounds like your CEO.
Key takeaways
- Voice cloning builds a synthetic copy of a specific person's voice from recordings of them speaking.
- Instant cloning tools work from one to two minutes of audio, and most commercial services rely on self-declared consent.
- Attackers use voice clones in vishing calls and voice messages to impersonate executives, officials and IT staff.
- A familiar voice is no longer proof of identity. Verification has to happen on a second, trusted channel.
- Realistic vishing simulations let teams practise that verification before a real attacker calls.
What is voice cloning?
Voice cloning is the use of AI to reproduce a specific person's voice, including pitch, accent, rhythm and timbre, so that new speech sounds like them. The output is a voice clone. It can read any text aloud or convert an operator's live speech into the target's voice during a call.
Voice cloning differs from standard text-to-speech, which uses generic stock voices that belong to no one in particular. It also differs from a video deepfake, which imitates a face. In security, a voice clone matters because it borrows the trust people place in a voice they recognise.
How does voice cloning work?
A voice clone is built in three stages: collecting audio of the target, conditioning an AI speech model on that audio, and generating new speech. The audio usually comes from public sources such as conference talks, earnings calls, podcasts, social media videos or a voicemail greeting.
The amount of audio required is small. In its instant voice cloning documentation, ElevenLabs recommends one to two minutes of clear audio, with a minimum of 10 seconds in total, and 30 minutes or more for a professional-grade clone.
Access controls are thin. Consumer Reports assessed six voice cloning services and found that four of them, ElevenLabs, Speechify, PlayHT and Lovo, relied on a checkbox in which users confirm they have the legal right to clone the voice, as The Register reported in March 2025.
How is voice cloning used by cybercriminals?
Cybercriminals use voice clones to make vishing calls and voice messages sound like someone the victim already trusts. The cloned voice carries the request: approve a payment, reset a password, share an MFA code or move the conversation to another app.
Common enterprise scenarios include:
- An executive asking finance to release an urgent transfer.
- An employee calling the IT help desk to reset credentials or enrol a new MFA device, a pattern covered in our analysis of AI voice phishing against IT help desks.
- A senior figure leaving a voice message to build rapport before sending a malicious link.
Voice clones are often paired with phone number spoofing, so the caller ID and the voice both appear legitimate. In sectors where a phone call can still approve a payment, the exposure is higher, as our guide to deepfake fraud in financial services explains.
What does a real voice cloning attack look like?
In a May 2025 public service announcement, the FBI warned that since April 2025 malicious actors had been sending text messages and AI-generated voice messages that claimed to come from senior US officials. The targets were largely current or former senior federal and state officials and their contacts.
The messages were designed to establish rapport first. The attackers then asked targets to move to another messaging platform through a link that harvested login credentials. The FBI's advice was direct: when a message claims to come from a senior official, do not assume it is authentic.
How can organisations defend against voice clone attacks?
The most reliable defence is procedural: a recognised voice should never be enough to authorise a sensitive action. Verification has to happen through a channel the caller does not control.
- Call back on a number from the company directory, never on the number that called or one given during the call.
- Require a second approver or a ticket for payments, password resets and MFA changes requested by phone.
- Agree on verification questions or code words for executive requests. In July 2024, a Ferrari executive ended a cloned-voice call impersonating CEO Benedetto Vigna by asking which book Vigna had recently recommended, as Fortune reported, citing Bloomberg.
- Treat audio cues as a bonus, not a control. The FBI lists odd tone, word choice and call lag as possible signs of a clone, but a good clone may show none of them.
- Train employees with realistic calls, so the callback reflex is practised before it is needed.
What terms are related to voice cloning?
| Term | How it relates to voice cloning |
|---|---|
| Vishing | Voice phishing. Voice clones are one of the tools attackers use to make vishing calls convincing. |
| Phone number spoofing | Falsifying caller ID. Often combined with a voice clone so both the number and the voice look familiar. |
| Synthetic media | The wider category of AI-generated audio, images and video. Video deepfakes of faces belong here rather than in voice cloning. |
| Social engineering | The manipulation of people into unsafe actions. Voice cloning is a delivery method for it. |
How does Arsen help teams prepare for voice clone attacks?
Arsen's vishing simulation uses real-time voice AI and voice cloning to run realistic vishing calls, including executive impersonation scenarios. Calls are generated dynamically rather than read from a fixed script, and voice, tone, urgency and tactics can be adapted to your organisation.
Security teams follow employee responses in real time and export results in PDF, CSV or through the API, which shows who verified, who complied and where procedures need reinforcing.