Sora Video Generator Copyright Controversy Escalates: Artists Accuse OpenAI of Plagiarism, Calls for Training Data Transparency Surge

News Lead

Recently, OpenAI's revolutionary video generation model Sora has once again become the focus of attention, but this time not for its stunning effects, but for being embroiled in an intense copyright dispute. Multiple artists have publicly accused Sora-generated videos of directly plagiarizing their original works, with lawyers, creators, and AI supporters engaging in heated debates on X platform. This incident has not only ignited discussions about intellectual property but also highlighted the urgency of AI training data transparency, drawing high attention from the global tech community.

Background

Sora is a text-to-video generation model released by OpenAI in February 2024, capable of generating high-quality videos up to 60 seconds long based on user text descriptions. With realistic imagery and precise physics simulation, it shocked the industry upon launch and was hailed as the "ChatGPT of video." However, its foundation relies on training with massive amounts of video and image data, and the sources of this data have become the focal point of controversy.

Even before Sora's release, copyright issues with AI-generated content were common. In 2023, image generation tools like Stability AI and Midjourney faced class-action lawsuits due to similar accusations. As a newcomer in the video field, Sora's training dataset is massive (OpenAI disclosed it exceeds millions of hours of video), but the specific sources have not been fully disclosed, setting the stage for this controversy.

Core Content

The dispute originated from a series of revelations on X platform this week. Independent artist Li Wei (pseudonym) first posted, showing her 2022 short video art piece "Dream Fragments" alongside a Sora-generated video prompted with "dreamy city nightscape, floating fragments," revealing striking similarities: identical composition, color tones, dynamic trajectories, and even subtle lighting details.

Li Wei stated: "I've tracked at least 5 Sora output videos that directly copy my style and elements. This isn't coincidence; the AI model has 'memorized' my work." Similar accusations spread rapidly, with another animator Zhang Hao sharing a "signature ink wash style transition" appearing in Sora videos, directly pointing to his award-winning work.

The X topic #SoraCopyright quickly topped trending searches with over 50 million views. Law firms initiated online petitions demanding OpenAI disclose its complete training dataset and assess infringement risks. OpenAI has not officially responded, but insiders reveal the company is reviewing relevant cases.

Various Perspectives

The creator camp is strongly dissatisfied. Well-known digital artist @ArtByEcho posted on X: "AI isn't magic; it consumes our labor without asking. This is plundering the artistic ecosystem!" She calls for establishing an "AI content watermark" mechanism to ensure generated content is traceable.

'Sora's power comes from collective intelligence, but if built on theft, everything will collapse.' — @ArtByEcho, X post, over 100,000 views.

Legal opinions are divided. Intellectual property lawyer Wang Lei (@IPLawExpert) analyzes: "Existing laws like the US DMCA struggle to address AI 'style imitation.' If training data includes copyrighted works, even 'transformative use' might constitute indirect infringement. The EU AI Act already requires high-risk models to disclose data sources, and China's 'Interim Measures for the Management of Generative Artificial Intelligence Services' also emphasizes compliance."

AI supporters counter-argue. Former OpenAI employee and current independent researcher Liu Yang states: "Sora uses publicly available internet data for training, similar to human learning processes. Perfect replication is rare; most are generalization results."

'Accusations need evidence chains: proving data was directly copied rather than learned patterns. Otherwise, it's a misinterpretation of 'style generalization.'' — Liu Yang, X live discussion.

OpenAI's official blog reiterates: "We respect intellectual property, actively remove infringing data, and explore licensing agreements." But critics argue this avoids the core issue: the black box of datasets.

Impact Analysis

This controversy has profound impacts on the AI industry. First, OpenAI's stock price experienced short-term volatility, with partners like Microsoft facing pressure. Second, it accelerates global legislative processes. The US Congress is reviewing the "NO AI FRAUD Act," aimed at prohibiting AI from using artists' works without permission; the EU's GDPR extension already treats AI data as "personal derivative rights."

For Chinese companies, ByteDance's Sora competitor VACE, Alibaba's Tongyi Wanxiang, and other models face the same challenges. Industry predictions suggest future training data will shift toward "licensed cleaning," with costs potentially rising 20%-50%. The creator ecosystem may be reshaped: more shifting toward NFT or paid licensing models.

Long-term, this incident promotes consensus on "responsible AI." Stanford AI Center Director Fei-Fei Li comments: "Transparent data is the ethical bottom line; otherwise, innovation will be hindered." The X topic #AITrainingTransparency has emerged, calling for open-source dataset standards.

Conclusion

The Sora copyright controversy is not just a technological collision but a turning point in the intellectual property paradigm of the AI era. If OpenAI can lead by publicly auditing its data, it might become an industry benchmark; otherwise, a wave of litigation will surge. Balancing innovation and rights tests the wisdom of the entire industry. As generative AI penetrates fields like film and advertising, transparent and fair rules urgently need to be established. As technology advances, we must not forget the humanistic bottom line.