YouTube likeness detection now covers voice: what Sume's API does

If you make YouTube videos with generated speech, the rule that matters is simple: do not copy a real person's voice without their agreement. The air.io summary of YouTube's 2026 changes, read on 2026-10-02, says YouTube extended likeness detection to speaking voice in August 2026. This post says what Sume's API does and does not do for voice, and what to keep on file.

What likeness detection checks and how a claim works are YouTube questions; the summary does not say, and neither does this post.

What does Sume's API do for voices?

Sume's public API does not clone a person's voice from a sample. Voice cloning exists in the Sume app only, not in the API, so a script that calls the API cannot create a voice that sounds like a named person.

What the API does offer for speech is text to speech through the TTS endpoints, speech to text, and avatar videos with Avatar 1.0, which is English only. Which voice a job uses is a request parameter you can log.

What should I log for every voiced video?

Keep a line per video: the script, the job id of the speech or avatar job, the voice setting you asked for, and the date. If you use a recording of a human, keep their written agreement and the original file.

The hosted MCP tools describe themselves, so an agent can read the exact fields: call tools_schema for the tool and keep the response with the record.

What about a voice that sounds like someone famous?

Do not prompt for it. A voice resembling a real person is the case likeness rules are written for, and a disclosure label does not fix it. Use a voice you have the right to use, or a person who consented.

What should I do now?

Check which of your channels use generated or recorded voices, write the consent or source next to each, and read YouTube's current help page on likeness detection before you publish the next batch.