YouTube is fighting deepfakes with new Likeness Detection tool


In an age of fragmented news, deepfakes present a real problem. YouTube is taking action to stop AI-generated videos being used to impersonate people.
The company has revealed details of its new Likeness Detection tool. For now, YouTube is targeting creators, giving them access to the tool – not only because creators are among those likely to be deepfaked, but also because the company will be hoping to be able to use their influence to hype up Likeness Detection.
This is not the first we have heard about Likeness Detection by any means; YouTube shared some details last month, but has hinted that it was working on such a tool for a while.
You can check out YouTube’s sneak preview of Likeness Detection in the video below:
Accessible via YouTube Studio, Likeness Detection requires anyone who wants to avoid their likeness being misused in deepfakes to upload facial scans along with copies of their government ID.
In terms of detecting deepfakes, YouTube will be doing all of the heavy lifting for those who sign up to the program. It will automatically analyze videos that are uploaded to the platform, running faces included in videos against the database of scans.
YouTube will issue warning messages when it believes someone has been deepfaked, and it will then be down to the individual to review the material and report the offending content for removal.
All of this will already raise alarm bells for some. There are privacy concerns, of course, but there is also the question of how people without recognized forms of ID will be protected, or people for whom the ID is different to their real world or online presentation.
For now, Likeness Detection is only available to people who are signed up as members of the YouTube Partner Program. YouTube has not given any information ab out when it may be more widely available, or provides details about whether or not prominent celebrities will be automatically protected without the need to manually upload images of their own.