Can't Use Copyrighted Characters in Sora Anymore and People Are Freaking Out (gizmodo.com)

OpenAI reversed course on Sora 2’s initial “copyright-free-for-all” approach, changing from an opt-out to an opt-in policy that requires rightsholders to explicitly allow their characters and content to be used in generated videos. The move follows a wave of problematic, often subversive recreations flooding the TikTok-style Sora app and public pressure from industry groups like the Motion Picture Association, which argued OpenAI must prevent infringement on the service. CEO Sam Altman framed the change as giving creators “granular control” while acknowledging edge cases and the need to iterate on enforcement. The shift matters for AI/ML because it highlights two fault lines: model training and content moderation. Reports indicate Sora’s capabilities stem from heavy use of copyrighted material during training, enabling highly accurate recreations—raising renewed legal risk even as OpenAI pushes for broad “fair use” treatment of model training. Technically, enforcement will require robust detection and filtering systems to map generations to licensed characters and block disallowed outputs, a nontrivial engineering challenge with false positives/negatives. The move also follows costly industry precedents (e.g., Anthropic’s settlement) and signals that platforms will need proactive rights-management and governance to balance creator control, legal exposure, and the user community’s appetite for remix culture.