Particle.news
Download on the App Store
64 ARTICLES
·
2h ago

OpenAI’s Sora 2 Goes Viral as Harmful and Copyrighted Clips Expose Safeguard Gaps

The invite-only iOS feed has raced to the top of the App Store, underscoring how Sora’s watermarks, metadata and cameo controls trail the risks of highly realistic, easily shared AI video.

The Sora app uses AI to generate fun, very realistic videos from user prompts.
Image
Nintendo characters Mario and Pikachu.
A vandalized ad from AI startup Friend in the New York City subway. Locals have targeted the campaign to denounce what they call “AI surveillance.”

Overview

  • OpenAI released Sora 2 alongside a TikTok-style Sora app that generates short, high‑definition videos with synchronized audio from text prompts and optional user likenesses.
  • Despite invite-only access, the app quickly hit No. 1 in Apple’s US App Store as Sora-made memes and highly realistic clips spread across larger social platforms.
  • Early use surfaced alarming material, including fabricated crimes, violent and racist scenes, and news-style footage from places like Gaza, Myanmar and Charlottesville, indicating safety filters were insufficient.
  • Copyrighted characters flooded the feed, and OpenAI confirmed rights holders must submit disputes to block uses rather than relying on a blanket opt‑out, a stance legal experts say is unlikely to hold up.
  • OpenAI touts moving watermarks and invisible metadata, but acknowledges the metadata can be stripped and is not a silver bullet, while outside researchers say these provenance measures need independent testing.