TikTok to label AI-generated content from OpenAI and elsewhere

TikTok plans to start labeling images and video uploaded to its video-sharing service that have been generated by artificial intelligence (AI), it said on Thursday, using a digital watermark known as Content Credentials.

Researchers have expressed concern that AI-generated content could be used to interfere with U.S. elections this fall, and TikTok was already among a group of 20 tech companies that earlier this year signed an accord pledging to fight it. The company already labels AI-generated content made with tools inside the app, but the latest move would apply a label to videos and images generated outside of the service.

TikTok's shift in policy is part of a broader attempt in the technology industry to provide more safeguards for AI usage. In February, Meta announced that it was working with industry partners on technical standards that will make it easier to identify images and eventually video and audio generated by artificial intelligence tools. Users on Facebook and Instagram would see labels on AI-generated images.

"We also have policies that prohibit realistic AI that is not labeled, so if realistic AI (generated contents) appears on the platform, then we will remove it as violating our community guidelines," Adam Presser, head of operations and trust and safety at TikTok, said in an interview.

The company said that the technology can attach metadata to content, which it can use to instantly recognize and label AI-generated content. TikTok said it began to deploy the technology on Thursday on images and videos and will be coming to audio-only content soon.

The Content Credentials technology was spearheaded by the Coalition for Content Provenance and Authenticity, a group co-founded by Adobe, Microsoft and others, but is open for other companies to use. It has already been adopted by the likes of ChatGPT creator OpenAI. YouTube, owned by Alphabet's Google, and Meta Platforms, which owns Instagram and Facebook, have also said they plan to use Content Credentials.

TikTok said it's the first video-sharing platform to put the credentials into practice and will join the Adobe-led Content Authenticity Initiative to help push the adoption of the credentials within the industry.

"TikTok is the first social media platform to support Content Credentials, and with over 170 million users in the United States alone, their platform and their vast community of creators and users are an essential piece of that chain of trust needed to increase transparency online," Dana Rao, Adobe's executive vice president, general counsel and chief trust officer, said in a blog post.

For the system to work, both the maker of the generative AI tool used to make content and the platform used to distribute the contents must both agree to use the industry standard.

When a person uses OpenAI's Dall-E tool to generate an image, for example, OpenAI attaches a watermark to the resulting image and adds data to the file that can later indicate whether it has been tampered with. If that marked image is then uploaded to TikTok, it will be automatically labeled as AI-generated.

TikTok, which is owned by ByteDance, has 170 million users in the U.S., which recently passed a law requiring ByteDance to divest TikTok or face a ban. TikTok and ByteDance have sued to block the law, arguing it violates the First Amendment.

(With input from agencies)