Can we use AI voices safely?

Where synthetic voices help

A synthetic voice is speech produced by AI from text. A cloned voice is a synthetic voice trained to sound like a particular person. Used well, they save real time:

• Training: narrated courses and safety briefings that can be updated by editing the script, not re-recording.
• Accessibility: documents and intranet pages read aloud for people with visual impairments, dyslexia or tired eyes.
• Languages: the same internal message voiced in several languages, checked by a native speaker before release.

Field Guide 07 covers voice and other media in AI tools; Field Guide 38 looks at one voice platform in practice.

Would your finance team spot a convincing call in your chief executive's voice asking for an urgent transfer? Book a free AI Value Briefing and we will look at it with you.

Consent comes first

A voice identifies a person, so treat a voice clone like a photograph or a signature. Before cloning anyone's voice, including your own chief executive's:

• Get written, informed consent that names the purpose, where the voice will be used, for how long, and how the person can withdraw.
• Agree what the cloned voice may never say: no instructions to pay, to share passwords or to change bank details.
• Store voice samples and the clone like other sensitive personal data, with limited access (Field Guide 10).
• Choose a provider with visible safeguards. ElevenLabs, for example, describes verification before its professional cloning tool, blocks on cloning celebrity and other high-risk voices, a tool to detect audio made on its platform, and use of the C2PA content provenance standard.

Where staff voices are involved, involve HR and your data protection lead before the first recording.

Tell people when a voice is synthetic

Listeners trust a human voice. Saying plainly that a voice is AI-generated protects that trust, and in the European Union it is also a legal matter. Under Article 50 of the EU AI Act, as shown on the European Commission's AI Act Service Desk:

• Providers of AI systems that generate synthetic audio must mark the output in a machine-readable format so it can be detected as artificially generated (Article 50(2)).
• Deployers, meaning organisations that use such a system, must disclose when audio, image or video content they produce is a deep fake; for clearly artistic or fictional work the duty is lighter (Article 50(4)).
• AI systems that talk directly with people, such as a voice bot, must be designed so people are told they are dealing with AI, unless that is obvious (Article 50(1)).

Check the current text and application dates at the official source, as the Act has been amended. A short spoken or written line such as "This narration uses an AI-generated voice" is a sensible habit everywhere. This is general information, not legal advice.

Voice cloning fraud and impersonation

The same technology helps criminals. In the United States, the FBI has warned that criminals use AI-generated audio to impersonate public figures or people close to the victim to obtain money, and to gain access to bank accounts (IC3, December 2024). A later FBI warning describes voice messages impersonating senior US officials and notes that AI voice clones can sound nearly identical to the real person (IC3, May 2025).

For an organisation, the classic case is the urgent call or voice note that sounds like the chief executive or finance director, asking for a payment, a change of supplier bank details or a password. The voice is no longer proof of identity.

Controls that help stop voice scams

The FBI's advice translates well to the workplace:

1. Hang up and call back on a number you already hold, never one given in the message.
2. Two people approve any payment or bank-detail change above a set amount, whoever seems to be asking.
3. Agree a code word or a known question for urgent requests among senior leaders and their assistants.
4. Never share passwords or one-time codes by phone, voice note or chat.
5. Limit public voice samples of executives where practical, and brief them that their voice can be copied.

Make it safe to say "I will call you back" to a senior person. A five-minute delay costs little; a fraudulent transfer can cost a great deal. Field Guide 26 covers wider guardrails for AI use.

Goes deeper in our free AI Field Guide: 7. Images, voice and files: giving AI more than text, 10. Data safety: what never to paste, and other rules, 26. Guardrails and permissions: what AI may and may not do, 38. ElevenLabs: AI voice, done responsibly.