TikTok Rolls Out Deepfake Likeness Detection Globally

AI security illustration: a shield sealing a red crack in a network

On November 19, TikTok announced two new tools for AI content. The first is Likeness Detection, which is starting a gradual global rollout. Verified creators and public figures will get an alert when someone has made AI content that uses their likeness, and they can report it directly. The second is a public lookup tool: anyone can upload an image or video to check whether it was generated or edited with TikTok's own AI features, and no TikTok account is needed.

Likeness Detection works in three steps. Eligible users join after a third-party provider verifies their ID document. When the system finds content or accounts that may be using their likeness, it sends a notification, and the user reviews each item in the tool. If something looks wrong, they report it straight to the safety team. TikTok says videos that break its Community Guidelines on harmful or misleading content, or that depict a person without consent, will be removed, and that the company does not keep users' ID documents. The tool covers government, politician and political party accounts as well.

The timing points to Europe's 2027 elections

TikTok presented the two tools at a safety and innovation event in London. Francesca Scapolo, the company's government relations lead for European elections, said:

"Protecting the integrity of the elections is one of the most critical responsibilities that we have at TikTok."

According to The Straits Times, several European countries including Italy, France and Greece are due to hold elections in 2027, and the EU began rolling out AI transparency rules in August that require AI-generated content such as deepfakes to be labeled. TikTok has itself required creators to label realistic AI images, audio and video for several years.

The lookup tool is meant to close a gap in labeling. In 2024, TikTok became the first video platform to adopt C2PA Content Credentials, so content made with its own tools automatically carries this provenance note, which other platforms can read to identify it. But once a video is downloaded, reposted or edited, the credential is often lost. Besides reading C2PA data, the new tool also detects VeraDura, TikTok's own invisible watermark, which the company describes as more durable and harder to remove. The lookup tool sits in TikTok's Transparency Center.

TikTok also released several figures. The platform has so far labeled more than 5 billion AI-generated videos. More than 99% of violating content is removed before anyone reports it. The in-app AI literacy hub launched in July has helped over 4 million people learn to spot AI content. And tools such as “Not interested” and keyword filters were used by more than 79 million people in the EU in the first half of this year.

A half step behind YouTube

On likeness detection, YouTube got there earlier. This year it opened a similar tool to celebrities and talent agencies, without requiring them to have a YouTube channel, and extended it globally around May. TikTok's version only began testing with a small group of US creators in July, with no timeline for expansion at the time. Four months later it is moving to a gradual global rollout.

The two approaches are similar: both require identity verification first, then face matching, with the platform acting on reports. The difference is scope. YouTube opened the door to public figures who are not on its platform, while TikTok's wording this time is “eligible creators and public figures.” Beyond its Help Center, it has not publicly spelled out how many followers or which account types qualify.

The lookup tool has narrower limits. It can only recognize content made with TikTok's own AI features, relying on credentials and watermarks that TikTok writes in itself. A deepfake generated with another company's model and then uploaded to TikTok will not be caught by this tool, so that content still depends on platform moderation and user reports.

Several questions remain unanswered by TikTok: the accuracy and false-positive rate of its likeness matching, how many countries the first wave covers, and how long it takes to resolve a report. With the European elections more than a year away, regulators and campaign teams are likely to want those numbers most.

Sources: TikTok Newsroom (workflow, eligible users and platform statistics for both tools), The Straits Times, AFP, CocoLoop, Tubefilter (scope of YouTube's likeness detection), Android Headlines; TikTok Help Center (statement that ID documents are not retained).