OpenAI's latest AI video generator Sora appears to have been trained on video game content and Twitch streams, raising potential legal concerns around copyright and intellectual property rights.
While OpenAI has not disclosed its complete training dataset for Sora, testing reveals the AI can generate videos resembling popular games and streaming content. The system can create clips similar to Super Mario Bros., first-person shooters like Call of Duty, and even recreate the likeness of well-known Twitch streamers.
OpenAI has implemented filters to prevent direct recreation of trademarked characters. However, creative prompting can still result in game-like content generation. The company has only acknowledged using "publicly available" data and licensed content from stock media libraries for training.
Legal experts warn this could be problematic. "Training a generative AI model generally involves copying the training data. If that data is video playthroughs of games, it's overwhelmingly likely that copyrighted materials are being included in the training set," explains Joshua Weigensberg, an IP attorney at Pryor Cashman.
The complexity increases with games like Fortnite, which involve multiple layers of copyright - the game developer's rights, the player's unique gameplay video, and user-generated content within the game. Each layer represents potential legal exposure.
Game content poses unique challenges due to proprietary elements like textures and character designs. While AI companies often claim fair use protection, the gaming industry's multi-layered IP rights make this defense more complicated.
Major game studios have remained largely silent on the issue. When contacted, most declined to comment or provide official statements about their stance on AI training using their content.
The situation becomes more complex when considering OpenAI's potential future applications. As the company explores "world models" that could generate interactive experiences, the legal implications of training on game content become increasingly relevant.
"Training an AI platform on the voices, movements, characters, songs, dialogue, and artwork in a video game constitutes copyright infringement, just as it would if these elements were used in other contexts," notes Avery Williams, an IP trial lawyer at McKool Smith.
As AI technology continues to advance, these legal questions surrounding training data and intellectual property rights will likely shape the future development of systems like Sora.
Note: I only included one link as it was the only one that could be contextually integrated while maintaining the article's structure and meaning. The other provided links were not directly relevant to the topic of AI video generation and gaming content rights.