YouTube Rolls Out Likeness Detection to Help Creators Fight AI Deepfakes
It gives creators a way to spot and remove fake videos before they spiral out of control.
A few years ago, AI videos were a fun internet trick, deepfakes of Tom Cruise doing magic tricks or Morgan Freeman narrating absurd monologues. Now, they’ve turned into something darker. According to DeepStrike, deepfake videos are expected to explode from 500,000 in 2023 to a projected 8 million by 2025, spreading faster than platforms can contain them. And with global AI adoption projected to hit 378 million users next year, according to AltIndex, the flood of synthetic content is only beginning.
YouTube, home to billions of videos and creators whose faces are practically brands, has become ground zero for the problem. Some creators have found AI versions of themselves promoting dodgy products or saying things they never said.
Even lawmakers around the world have started paying attention, alarmed by the potential for disinformation and identity abuse. The irony is that Google’s own AI tools, the same ones powering its creative breakthroughs, have helped make the mess worse.
YouTube is finally trying to clean up after itself. In 2024, reports surfaced that YouTube was developing two new safety tools: one to detect deepfakes of public figures and another called synthetic-singing identification technology, designed to spot AI-generated voices used without permission.
Now, the company is taking thing further and has begun rolling out a likeness detection system, a new feature that lets creators track and remove AI-generated videos that mimic their faces or voices. It’s basically Content ID for people’s faces, a system that scans uploads, spots possible fakes, and gives verified creators a say in what stays online.
Louis Eriakha
To get access, creators have to prove who they are by submitting a photo ID and a short selfie video. Some find that intrusive, but YouTube says it’s necessary for accurate results. Even then, the system isn’t perfect. It might flag parody or fair-use clips alongside harmful fakes, and reviewers still have to decide which videos truly cross the line.
The process feels rough around the edges, especially compared to how quickly copyright takedowns work. Creators have long been able to file DMCA takedowns within minutes, but AI likeness abuse is murkier. What actually counts as “misuse”? Besides, a satire clip with an AI-generated voice might stay up, while a convincingly fake brand endorsement will almost certainly go down. It’s a messy middle ground, one that depends as much on human review as it does on machine detection.
What’s clear is that YouTube is stepping into uncharted territory. It’s the first platform of its scale to give creators direct tools to fight AI impersonation, something competitors like TikTok and X are still fumbling with. And with Google’s new Veo 3.1 model soon integrating directly with YouTube, the company knows the wave of AI videos is about to get much bigger.
This new system won’t solve everything, but it’s a sign that YouTube knows what’s at stake. In a world where anyone’s face can be copied in seconds, giving creators a bit of control back might be the most human thing the platform can do.
Oyinebiladou Omemu
