TikTok takes on AI fakes and phonies

Like an angry punk rocker yelling at poseurs, TikTok has a plan to call out all (AI) fakes.

What happened: TikTok is poised to become the first social media platform to automatically label AI-generated images and videos on its app. Over the next few months, permanent digital watermarks will be added to such content, with the aim to expand to audio-only posts.

How it works: TikTok is using the Content Credentials system, developed by Microsoft and Adobe, which embeds metadata into AI content made by companies that voluntarily join the C2PA alliance. The group has over 100 members, including OpenAI, which joined this week.

Why it matters: Scoping out AI-generated content is tough — take the Met Gala deepfake quiz we ran this week, in which 57.3% of respondents were fooled by an AI fake — and concerns over AI-fuelled disinfo have hit new heights amid the busiest election year ever. 

  • A fake recording of a Slovakian presidential candidate saying he rigged the election and deepfakes of Bollywood stars criticizing Indian PM Narendra Modi are just two prime examples of how AI has meddled with high-profile elections this year.

Yes, but: The Content Credentials system TikTok uses is opt-in, so content made with AI platforms that haven’t joined will remain unmarked. Plus, there are ways disinformation agents could get around the system, like manually removing metadata.—QH