Data, AIGC & Copyright, Allen's Speech at AIPPI World Congress

On September 14, 2025, Allen Wang, Chief Partner of Beijing TA Law Firm, Chairman of Copyright Committee of AIPPI China Group, delivered an English speech at AIPPI (International Association for the Protection of Intellectual Property) World Congress held in Yokohama, Japan. The detailed transcript is provided below.

After listening to the presentations from our European and U.S. colleagues, I must say the feeling of “chaos” becomes even stronger. From their cases and legislative updates, we did not see a clear and unified direction emerging.

From the European side, we heard about the TDM exceptions, but also that these were never designed with AI in mind. From the U.S. side, Fair Use is being tested case by case, and the national AI strategy deliberately avoids copyright questions. So far, no consistent standards are visible, and no single direction is guiding us.

In China, we also face many challenges. I would like to share four aspects of our current issues and reflections.

Part I: Authorship & Originality

The first challenge is authorship and originality. In China, we already had one case — the text-to-image case — where the court recognized copyright because the human input was sufficiently specific. But in the text-to-video case, the court has not issued a decision. They show a much more cautious attitude.

The real issue is how we define and evaluate the “human intelligence contribution.” It is not enough just to say a person typed in a self-created prompt. We need to ask: did the human exercise creative control over the expression? Did they refine prompts, select and revise the outputs, edit, combine, or supervise the process? These actions can show expressive control. The big open question for all of us is: at what point do these actions cross the threshold into authorship?

Part II: Training Data Use

The second challenge is training data. In China, our Copyright Law has a closed list of fair use exceptions — things like teaching, research, and private study. Similar to EU, these provisions were never designed with AI training in mind. Under our current law, it is very difficult to say that AI training is covered.

Another unique point is why we haven’t seen major public cases about Data training in China yet. Many platforms are still “quietly training and learning” — using data in the background without pushing disputes into court. And structurally, many of the investors and developers of large AI models in China are themselves the biggest copyright holders of traditional works — music, texts, images, audiovisual contents, and network dissemination rights. They are both the builders of AI and the right holders of massive human-created content. That’s why open conflicts haven’t erupted in China. But this silence is only temporary. Sooner or later, the use of such data will have to be tested in court.

Part III: Style Imitation

The third challenge is style imitation. Artistic works are often closely tied to the artist’s own style, which gives them strong recognizability and unique commercial value. Some artists, like the Japanese animation artist, Hayao Miyazaki, have such distinctive and pioneering styles that their works are instantly recognizable.

In the age of AI training and generation, these highly distinctive styles are being massively imitated and widely spread. If we say that style itself, as an idea or creative concept, is not protected by copyright law, then the question is: how do we protect the creative style of artists like Miyazaki?

The reality is that when AI produces endless outputs in a particular style, the original distinctiveness, the market recognition, and even the public’s aesthetic experience can be diluted. People may even develop aesthetic fatigue. The result is that the artist’s original competitive advantage and market value are weakened or eroded. So perhaps we need to look beyond copyright law and ask: can unfair competition law play a role in protecting such distinctive artistic styles against commercial free-riding?

Part IV: Enforcement

The fourth challenge is enforcement, especially across borders.

Imagine a situation where in Country X, AI-generated content is recognized as a copyrightable “work”. But once that same content crosses into Country Y, it is not protected at all — or vice versa. How do we resolve licensing and conflicts in such cases?

Without at least some common ground on baseline rules, enforcement will remain fragmented, and global licensing will stay uncertain. This is not just a legal question — it directly impacts how the industry develops and how creators and companies can collaborate across borders.

Slide 2 – The First Text-to-Video Case in China

Let me turn to the very first text-to-video case from China that is still ongoing —. It is being heard by the Beijing Internet Court and has already gone through four full hearings, but no decision has been issued. This delay itself shows how cautious the court has become.

What is striking is that the same court previously recognized copyright in the first text-to-image case, where specific prompts and human input were seen as sufficient to establish human contribution. But in this text-to-video case, the court is far more hesitant. The hesitation reflects both the complexity of the issue and the strong academic opposition to granting copyright to AIGC outputs.

For the plaintiff, the facts are straightforward: the defendant’s video was almost a one-to-one copy. Yet before the court can even address infringement, it must first decide whether the AI-generated video qualifies as a “work” under copyright law. That is the real dilemma.

The plaintiff himself complained to me directly: “The outcome of this case will decide the future of the industry. If copyright is not granted, then the industry cannot survive. Even if someone signs a contract with me worth 100 million yuan, it means nothing — because I cannot deliver the copyright.” This shows how, for creators, the uncertainty of copyright is not just academic — it is a matter of survival.

Slide 3 – Scholarly Opposition

OK, the academic opposition. One important point is how much influence scholar’s voices really have. And here we see a difference between the U.S. and China. When I spoke with American colleagues, someone told me quite frankly that scholar’s opinions don’t matter in practice — no one really cares what the professors say, and judges will not base their rulings on academic debates. But in China, it is different. Most of our specialized IP judges, even at the Supreme Court, were once students of the most authoritative professors in this field. These professors are also frequently invited as expert advisors in legislation and major IP cases. So when leading scholars raise strong objections, their voices really do matter, and they make judges much more cautious in handling cases. I don’t know, Judge Reto, did your professor ever give you any pressure on this.

Some Chinese scholars have clearly argued against granting copyright to AIGC. Their main points can be summarized like this:

1. No human consciousness, no creation. Without human intent or will, there is no creation in the legal sense.

2. Prompts are ideas, not expression. They are instructions, not authorship.

3. Outputs are uncertain and uncontrollable. Users cannot claim “what you see is what you get.”

4. Cross-medium generation breaks human’s artistic logic. For example, text-to-image or text-to-music is unlike human artistic creation.

5. Algorithms and datasets dominate expression. Not human input.

6. International disharmony. If one country protects AIGC but another does not, this risks imbalance under the Berne Convention.

This explains why, in China, even when the facts look clear, courts hesitate. Academic opposition has a real effect on judicial practice in China.

And this brings us directly to the next challenge. If we only listen to academic theory, we might get one answer. But in reality, professional creators who actually use AI work in very different ways. The gap between academic “should be” scenarios and industry “as it is” practice is what I want to highlight next.

Slide 4 – The Reality of Professional Use vs. Academic Assumptions

The gap between theory and practice. Academics often describe AIGC as if it were “one-click” — you type in some text, and immediately get a finished video, image, or piece of music. But this is not how professional creators actually work.

In practice, professional use is much more complex, and human contribution is inevitable. The final expression of the work always requires human intervention, adjustment, and creative control. That can mean designing prompts over multiple rounds, refining outputs, editing material, or adding dubbing, scoring, and other artistic choices. We should not be misled by the label “text-to-video” — it is not just “text in, video out.” Human input in shaping the final result is indispensable. And the more professional the creator, the stronger their control, and the clearer their originality is reflected in the work.

I have spoken with several professional film directors who actively use AI tools in their production process. They told me that in their experience, the final videos generated with AI can reach a point where about 70% of the visual expression is under human control. In other words, the creative intention, the expressive adjustments, and the final look of the work are still very much directed by humans, not the machine.

We also need to remember that methods of creation have always been evolving. Take the right of cinematography: we moved from live-action film shooting to animated film production, and the scope of the right expanded. But human authorship remained at the center. Today, AI-assisted or human–AI collaborative creation maybe the latest stage in this evolution. It is not in conflict with copyright principles — it shows how copyright adapts as new technologies emerge.

Slide 5 – From Challenges to Solutions

So after laying out the challenges, let me end with a few suggestions on possible directions for solutions.

First, the idea of “wide-in, narrow-out” for training data.

At the input stage, AI training should be allowed broad access to high-quality datasets. This is necessary to optimize models and improve the quality of outputs. But at the output stage, stricter scrutiny is required. Only outputs that meet legal standards and respect authors’ rights should be protected. In this way, we balance innovation with accountability.

Second, we need to clarify what counts as meaningful human contribution.

We should define standards for creative control, managerial supervision, and editorial intervention. Clearer benchmarks will give guidance to courts, industry, and creators, and ensure more consistent judgments on authorship.

Third, courts need expert involvement.

In complex AI-related cases, it would help to invite technical experts and creative industry specialists as expert assistants to the court. Their expertise can shed light on how AI systems work, how much control humans exercised, and whether the output embodies sufficient originality.

And finally, we may need to think about new legal mechanisms.

As a civil law country, legislative reform in China takes time. It is very difficult to align different legal traditions across borders. That is why some have suggested exploring a sui generis regime for AI-assisted works. Such a framework could give international clarity on three key questions:

whether an AIGC output is to be regarded as a “work”;

if so, how it can enjoy copyright protection;

if not, whether it should be treated as a distinct type of content, with clear rules for licensing and remuneration.

This might be a more meaningful choice than forcing AIGC into the rigid boundaries of traditional copyright. At the very least, it would provide clarity for creators, industries, and users, and help us reduce conflict across borders.

That concludes my sharing, thank you all for your time and interest.

Provided in-depth assistance to clientele in the areas of generative AI, and advised global clients on cutting-edge issues in:

Generated Content in the fields of literature, film, animation, graphics, music, short videos etc
Foundation Models Development and Data Training
Generative AI & IP Cross-border Disputes
Data privacy and protection
AI regulatory compliance & Algorithmic Governance
AI ethics advisory & Risk Mitigation

Served as leading author of the China Chapter of the book Comparative Series: AI Generated Contents and Copyright Law of AIPPI, and frequently invited to participate in international conferences, academic seminars and intellectual property forums, where he delivered speeches on topics related to generative AI and AIGC.

Recent Speeches and Publications by Allen Wang in the Field of Artificial Intelligence

Speaker, Can Artificial Intelligence Claim Copyright Ownership? IPR Gorilla, Dubai, November 2019.
Author, Intellectual Property and Security Regulation of AIGC, 2022.
Speaker, Metaverse & Chinaverse, LESI, Montreal, Canada, April 2023.
Speaker, Copyright Protection of Industrial Design, AIPPI, Istanbul, October 2023.
Speaker, Exploring the Copyrightability and Ethical Regulation of AIGC from an International Perspective, Tsinghua University, Beijing, April 2024.
Speaker, Copyright Law Issues in the Training Process of Generative AI Data in China, AIPPI, Hangzhou, October 2024.
Speaker, Film and Television Content Creation and Copyright Attribution in the Era of AI Technology, China Golden Rooster and Hundred Flowers Film Festival, Xiamen University, November 2024.
Speaker, Copyright Protection of Different Types of Artistic Works and the Challenges Posed by AI-generated Creations, China Federation of Literary and Art Circles, Beijing, November 2024.
Instructor for the workshop on AI-Generated Musical Works, Graduate Program in Arts Management, Zhejiang Music Institute, Hangzhou, December 2024.
Speaker, Navigating New Horizons: Opportunities and Challenges in China's Audio and Visual Copyright Landscape, LESI webinar, February 2025.
Speaker, Opportunities and Challenges Brought by Artificial Intelligence to Literary and Artistic Creation and Copyright Protection, China Federation of Literary and Arts Circles, Hangzhou, April 2025.
Speaker, Generative AI: Avoiding Potential Copyright and Licensing Pitfalls, LESI, Singapore, April 2025.
Co-author, Copyrightability and protection of AIGC in Content Industry, China Internet Audiovisual Industry Development Report, 2025.
Leading author, the China Chapter of the book, Comparative Series: AI Generated Contents and Copyright Law, AIPPI, 2025.
Instructor for the workshop on AIGC by professional artists, China Federation of Literary and Art Circles, Beijing, June 2025.
Instructor for the Master Class of China TV Drama producer, Beijing, June 2025.
Project leader for the 2025 AIPPI global Study Question on Copyright and Artificial Intelligence in China, Yokohama, 2025.
Speaker, Data, AIGC & Copyright, fair use or infringement? AIPPI, Yokohama, 14 September 2025.
Speaker, Industrial Development and IP Protection driven by AI, Xian, 24 September 2025.
Speaker, AI & Copyright: what can be protected or infringed? LES Pan-European meeting, The Hague, 30 September 2025.

- END -

本栏目文章为本所为本行业及社会公众提供的公益性普法服务，不属于针对具体事项的法律意见，也不代表本所针对具体个案的意见或观点。