YouTube’s Deepfake Detector Will Start Listening for Your Voice Later This Year

OpenAI's Sora 2 and Google's own Veo models can turn a text prompt into a video with synchronized dialogue. YouTube's defense against fakes built that way has been a tool that watches uploads for your face. At Made on YouTube on Wednesday, the company said it will start listening for your voice, too.

The tool is called likeness detection, and it works a lot like Content ID, except the thing it hunts for is you. Later this year, YouTube says, it will begin "integrating speaking voice detection with facial detection to improve overall match accuracy." It's also bringing the tool to the YouTube mobile app, so you can enroll, get match alerts, and act on them from your phone. YouTube didn't give a date for the mobile version.

Who can sign up

YouTube's announcement says "billions of channels" are eligible. The rollout history explains the big number. Likeness detection started in December 2024 as a partnership with CAA, with early testing planned for the agency's clients, including actors and NBA and NFL players. In October 2025 it opened to a first wave of YouTube Partner Program creators. In March 2026 YouTube added a pilot for government officials, political candidates, and journalists, and in April it added talent agencies and the celebrities they represent. In May it opened to everyone 18 and older. You need to be an owner or manager of the channel, and YouTube says the tool isn't available in some countries.

What happens when it finds a fake

Matches show up in YouTube Studio under Content detection, then Likeness. For each flagged video you can file a likeness removal request through YouTube's privacy complaint process, file a copyright takedown if the video reuses your real footage, or archive it and leave it up. YouTube emails you the outcome. Removal isn't automatic. YouTube weighs whether a video is parody or satire, and it says plainly that detection doesn't guarantee removal.

Why podcasters should care about the voice part

Podcasters and commentary creators have the most riding on voice detection. Their voices are the product, their faces may barely appear on screen, and a cloned voice makes a convincing scam ad or fake hot take. Today the tool only matches faces, so if someone clones your voice, you have to find the video yourself and file a privacy complaint. YouTube's wording is careful. Voice gets folded into facial matching to improve accuracy, "laying the groundwork for broader voice protections over time." I read that as better face matches first and standalone voice-clone detection later. If your cloned voice turns up over somebody else's face, I wouldn't count on an alert this year.

What you hand over

Enrolling means giving YouTube a government ID and a selfie video. YouTube builds a template of your face from that selfie and your uploads, and for voice it says it will process audio from your videos and may ask for a voice reference later. It keeps the data for up to three years after your last sign-in, deletes it when you turn the feature off, and says it won't use it to train Google's generative AI models without your consent. A separate, optional checkbox lets YouTube use your face and voice templates to improve its detection models. You can leave it unchecked, or uncheck it later in Studio.

The law may eventually give you another option. In June the Senate Judiciary Committee advanced the NO FAKES Act by unanimous voice vote, sending it to the full Senate. For now, the practical protection is a biometric deal with the company whose parent also builds Veo. If your face or voice is how you make a living, that trade is probably worth making. Decide on the model-training checkbox on purpose.