As OpenAI recently released its video-generating AI model Sora, it has sparked widespread attention. An in-depth investigation has revealed that the model's training data may contain a significant amount of unauthorized game videos and live-streaming content, which could pose substantial legal risks for the company.
Through detailed testing, researchers found that Sora is capable of generating videos in various gaming styles, ranging from scenes reminiscent of "Super Mario Bros." to first-person shooter visuals akin to "Call of Duty," and even the typical style of 90s arcade fighting games. More impressively, Sora also demonstrated an astonishing understanding of Twitch live-streaming content, being able to create characters strikingly similar to popular streamers Auronplay and Pokimane.
Intellectual property lawyers have issued serious warnings regarding this matter. Joshua Weigensberg, a lawyer at Pryor Cashman, pointed out that using unauthorized video game content to train AI models could constitute a significant legal risk. Experts emphasize that copyright issues surrounding game videos are extremely complex, involving not only the content copyrights of game developers but also the unique rights of players and video creators regarding their game recordings.
OpenAI has maintained a vague stance on the sources of its training data. The company only acknowledges using "publicly available" data and licensed content from media libraries like Shutterstock, but has not specified the exact sources of game videos. This lack of transparency has heightened industry concerns about potential copyright infringements.
Currently, the tech industry is facing a series of similar intellectual property lawsuits. From Microsoft and OpenAI being accused of reusing licensed code, to AI art applications facing allegations of infringing on artists' rights, and music AI startups being sued by record companies, copyright issues have become one of the biggest obstacles in the development of generative AI.
Legal experts warn that even if AI companies might ultimately win these lawsuits, individual users could still face risks of intellectual property infringement. "Generative AI systems often produce recognizable intellectual property assets," Weigensberg stated. "Regardless of the programmer's intentions, complex systems can still generate materials that are protected by copyright."
As the technology of world models continues to evolve, this issue becomes even more complicated. OpenAI believes that Sora can essentially generate video games in real-time, and the similarity of this "synthetic" gaming to the training content could lead to further legal disputes.
Industry lawyer Avery Williams candidly stated: "Training AI platforms using sounds, actions, characters, songs, dialogues, and artworks from video games essentially constitutes copyright infringement." The legal disputes surrounding "fair use" will have profound implications for the video game industry and the creative market.
As one of the most notable AI video generation technologies currently, Sora's training methods reflect the legal gray areas faced by generative AI. In the ongoing struggle between technological innovation and intellectual property protection, OpenAI will encounter increasingly significant challenges.
