Sora 2 Copyright Issue: Legal Risks and AI Policy Shift
Oct 07, 2025UP!
Sora 2 Copyright Issue: Legal Risks and AI Policy Shift
In October 2025, OpenAI launched a new “Cameo” tool in its video model Sora 2. It lets users add real people’s faces and voices to AI videos. Within hours, clips with famous characters spread online.
Pikachu and other icons appeared across social media. CEO Sam Altman did not post them himself. Still, his decision triggered a worldwide copyright debate. On October 4, OpenAI added stricter content filters, showing a move toward responsible AI.
Three Legal Issues Around Sora 2
AI works in three phases: training, generation, and output. Japan’s Copyright Act Article 30-4 allows data use for training. But new risks appear once users share the output online.
Issue 1: When Does AI Output Become Reproduction?
Copying a known character’s look or voice likely counts as reproduction or adaptation.
Creating a Pikachu-style video is not data analysis. It borrows creative elements from the original work. Only the rights-holder can authorize that. Without consent, it’s likely copyright infringement.
Issue 2: Can Developers Be Liable?
If misuse is predictable and ignored, developers may share liability.
Japan’s Winny Case set two key standards. A developer is liable if they know about infringement or foresee it but take no action. With Sora 2, OpenAI could foresee misuse involving famous characters. That raised its duty of care.
Attorney Tsunoda’s View: OpenAI’s global scale increases its responsibility. The new filters show the company’s first step toward compliance.
Issue 3: Posting Online and Public Transmission
Uploading AI videos ends private use and violates public transmission rights.
Private copying is allowed only for personal use. Once shared online, it becomes public distribution. That breaks the creator’s transmission rights.
- Uploader: Direct infringer sharing the content.
- Provider: Liable if repeated misuse is tolerated.
Preventive Responsibility in AI
Altman’s October 4 statement marked a policy shift. AI firms must act before infringement occurs. Preventive design and early detection now define responsible AI.
FAQ: Sora 2 and Copyright
Q1. Can AI learn from copyrighted data?
A1. Yes. Article 30-4 allows data use for analysis. But copying the same look or voice is not allowed.
Q2. Can I post parody clips online?
A2. No. Private viewing is fine. Sharing them online violates public transmission rights.
Author
Akasaka International Law & Accounting Office
Attorney Shinji Sumida
You are welcome to contact us via the Contact Form to discuss and for more information.