Trust & safety

Moderation and appeals

What happens after somebody reports content made here, what we are able to do about it, and what to do if a decision went against you and you think it was wrong. Written plainly, because a rule you cannot see enforced is not a rule.

The honest shape of it: enforcement here is report-driven. There is no pre-generation content filter reading your scripts and no automated voice-matching that decides whether a sample is really yours — we do not have those, and a page that implied we did would be worse than useless to someone relying on it. What we have is a consent record you accept before generating, a reporting route that reaches a person, and the ability to act on the account behind the content.

What happens when something is reported

01A report arrives. From the report form, or by email. It is validated, written to a durable server-side record, and forwarded to our trust-and-safety address. The intake is rate-limited so it cannot be used to flood us, and every attempt is logged either way.
02A person reads it. Not a classifier. Reports go to a human queue, and the 48-hour target is a target for a person having looked at it — not for an automated acknowledgement email.
03We establish what we can. What was generated, by which account, and whether a consent acknowledgement exists against it. Every account accepts the Responsible Use policy before using the studio, and that acceptance is stored as a dated record, so the question “did this person assert they had permission” has an answer.
04We act, in proportion. Removing content we host, refusing a voice, suspending an account, or terminating it, depending on what happened and whether it looks deliberate. A first mistaken clone of a friend's voice is not the same thing as a fraud attempt, and is not treated as though it were.
05Both sides are told. The reporter hears what we did, if they gave us a way to reply. The account holder is told what was actioned and why, and how to appeal — an enforcement notice that does not explain itself is not enforcement, it is a disappearance.

What we can and cannot do

We can

Remove generated audio and voice profiles that we host.
Refuse or delete a cloned voice.
Suspend or terminate an account for breaching the Terms or the Responsible Use policy.
Tell you whether a given account had accepted the consent policy, and when — the record is dated.

We cannot

Remove a file from somewhere else. Once audio has been downloaded and published on another platform, that platform is who can take it down — we can confirm to them what we removed on our side.
Prove a voice is yours from a recording alone. We can act on a credible report; we cannot run a voice-identity match, and we will not pretend to.
Give legal advice, or decide a rights dispute between two people.
Act on content that was never made here. If it was not generated on VoiceLabs, there is nothing on our side to remove.

Appealing a decision

If content of yours was removed, a voice was refused, or an account was suspended and you believe that was wrong, say so. Appeals go to the same people, and we aim to acknowledge one within the same 48 hours we hold ourselves to for reports.

An appeal is far easier to act on with four things in it, so include them if you can:

What was actioned — the generation, the voice profile, or the account, and roughly when.
Why you think it was wrong— in your own words. “It is my own voice” is a complete answer, if it is true.
Any permission you hold — a message from the speaker, a release, a contract clause. It does not have to be a formal document.
The account email you used — so the appeal can be tied to the decision.

A decision can be reversed. If we got it wrong we restore what we can and say so; if we still think it was right after looking again, you will be told why rather than simply hearing nothing.

If your voice was cloned without permission

You do not need a VoiceLabs account to report it, and you do not need to prove anything before we will look. Use the report form, describe the content, and tell us where you encountered it. Audio generated here also carries an inaudible, machine-readable watermark, consistent with EU AI Act Article 50 — it is one signal among several for establishing that a file came from this service, and it is not something you have to find yourself before reporting.

Every report is read by a person, with a 48-hour triage target.

Report content Appeal a decision

See also Responsible Use for what is allowed and prohibited, Audio rights for what you may do with what you generate, and Terms section 14, which is the clause suspension and termination sit under. This page describes how we work; the Terms govern.