YouTube enhances AI deepfake defense for officials, journalists

YouTube has announced the expansion of its advanced likeness detection technology, designed to identify AI-generated deepfakes, to a select pilot group comprising government officials, political candidates, and journalists. Participants in this program will be granted access to a specialized tool capable of identifying unauthorized AI-generated content, empowering them to request its removal should they deem it in violation of YouTube's established policies.

This technology was initially rolled out last year to approximately 4 million creators within the YouTube Partner Program, following a series of preliminary tests.

Operating in a manner akin to YouTube’s established Content ID system, which identifies copyright-protected material in uploaded videos, this likeness detection feature specifically targets simulated faces generated by AI tools. Such tools are frequently exploited to disseminate misinformation and distort public perception by creating deepfaked personas of prominent individuals, including politicians and government officials, to portray them saying or doing things they never did in reality.

Through this new pilot initiative, YouTube endeavors to strike a crucial balance between fostering users' freedom of expression and mitigating the inherent risks posed by AI technology's capacity to generate highly convincing likenesses of public figures.

"This expansion is really about the integrity of the public conversation," stated Leslie Miller, YouTube’s Vice President of Government Affairs and Public Policy, during a press briefing preceding Tuesday’s launch. She further emphasized, "We know that the risks of AI impersonation are particularly high for those in the civic space. But while we are providing this new shield, we’re also being careful about how we use it."

Miller clarified that not every detected match would automatically result in removal upon request. Instead, YouTube will meticulously evaluate each submission in accordance with its existing privacy policy guidelines, specifically to ascertain whether the content qualifies as parody or political critique – recognized forms of protected free expression.

Furthermore, the company highlighted its active advocacy for these protections at a federal level, notably through its support for the NO FAKES Act in Washington D.C., legislation designed to regulate the unauthorized use of AI in creating recreations of an individual’s voice and visual likeness.

To utilize the new tool, eligible pilot testers are required to first verify their identity by submitting a selfie alongside a government-issued ID. Once verified, they can establish a profile, review any detected matches, and, if desired, request their removal. YouTube has indicated future plans to empower individuals with the capability to prevent the upload of violating content before it goes live, or potentially to enable monetization of such videos, mirroring the functionality of its Content ID system.

While the company refrained from confirming the specific politicians or officials participating in the initial testing phase, it affirmed its long-term objective is to make this technology broadly accessible.

Regarding the placement of an AI-generated content label, Amjad Hanif, YouTube’s Vice President of Creator Products, explained, "There’s a lot of content that’s produced with AI, but that distinction’s actually not material to the content itself." He elaborated, "It could be a cartoon that is generated with AI. And so I think there’s a judgment on whether it’s a category that maybe merits from a very visible disclaimer."

YouTube is not currently disclosing the precise number of AI deepfake removals facilitated by this detection technology among creators, though it did state that the volume of content removed to date has been "very small."

Hanif further commented, "I think for a lot of [creators], it’s just been the awareness of what’s being created, but the volume of actually removal requests is really, really low because most of it turns out to be fairly benign or additive to their overall business."

However, this trend may not hold true for deepfakes involving government officials, politicians, or journalists, where the implications could be significantly different.

Looking ahead, YouTube plans to extend its deepfake detection capabilities to additional domains, encompassing recognizable spoken voices and other forms of intellectual property, such as popular characters.