YouTube Fights AI Deepfakes Targeting Public Figures

YouTube Strengthens Its Technology Against AI Deepfakes

YouTube has announced the expansion of its technology for detecting deepfakes generated by artificial intelligence. This pilot program now includes government officials, political candidates, and journalists. These users will have access to a tool that allows them to identify AI-generated content created without their consent and request its removal if they believe it violates the platform's rules.

This technology, initially launched last year for around four million YouTube creators as part of the YouTube Partner Program, has been tested before being implemented on a larger scale.

A Tool Inspired by Content ID

The new similarity detection tool operates similarly to YouTube's Content ID system, which identifies copyrighted content in uploaded videos. This system searches for simulated faces created by AI tools, often used to spread misinformation by manipulating perceptions through deepfaked personalities of notable figures, such as politicians or other officials, making them say or do things they have not actually done.

With this pilot program, YouTube aims to balance users' freedom of expression with the risks associated with AI technology that can create convincing likenesses of public figures.

Preserving the Integrity of Public Discourse

Leslie Miller, YouTube's Vice President of Government Affairs and Public Policy, emphasized the importance of this expansion for the integrity of public discourse. She explained that the risks of AI impersonation are particularly high for those operating in the civic space. However, she clarified that not all detected matches would be automatically removed. YouTube will evaluate each request according to its privacy policy guidelines to determine if the content is a parody or political criticism, which are forms of protected free expression.

Support for the NO FAKES Act

YouTube is also supporting the NO FAKES Act in Washington, legislation that would regulate the use of AI to create unauthorized reproductions of an individual's voice and visual appearance.

To use the new tool, participants in the pilot program must first verify their identity by uploading a selfie and an official ID. They can then create a profile, review detected matches, and request their removal if necessary. YouTube is considering allowing users to prevent the uploading of rule-violating content before it goes live, or potentially to monetize these videos, similar to the Content ID system.

Towards Broader Availability

Although YouTube has not specified which politicians or officials will be the first to test this technology, the company has expressed its intention to make this tool widely accessible in the future.

AI-generated videos will be labeled accordingly, but the placement of these labels varies. For some videos, the label appears in the description, while for sensitive subjects, it is placed at the beginning of the video. This approach is similar to that adopted by YouTube for all AI-generated content.

Still Limited Use

Amjad Hanif, YouTube's Vice President of Creator Products, explained that while there is a lot of AI-produced content, this distinction is not always significant for the content itself. He noted that the volume of requests for the removal of AI deepfakes is currently very low, as most content turns out to be benign or adds to the overall activity of creators.

However, this could change with deepfakes involving government officials, politicians, or journalists. YouTube plans to expand its deepfake detection technology to other areas, including recognizable spoken voices and other intellectual properties like popular characters.