AI Search Becoming a “Piracy Carrier”? Different Legal Approaches to RAG Technology in China and the US
Hello everyone, I’m your financial journalist and economist. Today, we’re going to discuss a topic that seems technical but actually affects everyone’s daily habits—and more importantly, the fate of future internet giants: Should platforms compensate for pirated content found through AI search, especially with RAG (Retrieval-Augmented Generation) technology?
Recently, the Supreme People’s Court of China ruled on a significant case involving an AI search engine called “Mita.” When users searched for TV dramas, the engine directly provided links to pirated files stored on online file-sharing platforms. In the eyes of Chinese courts, this didn’t constitute infringement; however, from the perspective of the US Copyright Office, the situation might be much more complicated.
This isn’t just a matter of differing legal provisions; it reflects two distinct approaches to protecting industries. Let me break down the issue into five key points to help you understand the underlying issues.
---
Case Review: Is It Fine for a Platform to Quickly Remove a Pirated Link Found by AI?
First, let’s clarify the facts:
Main Players: Mita AI Search Engine (a company that provides AI search services) vs. the copyright holders of the TV dramas “Zhuangding Ye Shi Bing” and “An Hua.”
Events:
1. The copyright holders searched for these dramas on Mita.
2. The first result Mita provided was a link to a third-party file-sharing platform.
3. The video content was identical to the original, indicating it was pirated.
4. The copyright holders complained to Mita, and Mita immediately removed the link.
5. The copyright holders sued, demanding compensation.
Outcome: Both the first and second-instance trials in Shanghai courts dismissed the copyright holders’ claims, ruling that Mita had not infringed.
Why?
The court’s reasoning was straightforward, based on the well-known “Safe Harbor Principle”: As a platform, Mita is not the uploader; as long as it doesn’t actively distribute the content and removes it promptly upon receipt of a complaint, it is not liable.
- No Active Distribution: The link came from a third-party platform, and Mita merely displayed the search result.
- No Intent to Infringe: AI technology currently struggles to automatically identify pirated links. Mita had registered its algorithm and removed the link immediately, fulfilling its duty of “reasonable care.”
- Conclusion: Mita had no malicious intent, and its actions were timely, so it was deemed not responsible.
In Simple Terms: In current Chinese judicial practice, AI search engines are considered intermediaries. As long as they don’t engage in illegal activities and respond to complaints, the law protects them.
---
Chinese Courts’ Logic: Protecting Innovation, Giving AI Room to Experiment
Why did Chinese courts make this decision? There are underlying industrial considerations:
1. AI as a Carrier, Not a Creator: Courts believe that although RAG technology is used, AI still retrieves information from the internet; it doesn’t create new content. Therefore, traditional internet platform rules apply.
2. Technical Limitations as a Defense: Courts emphasized that current technology cannot automatically identify pirated content. This means AI isn’t responsible for pirated links since it can’t distinguish them. Requiring AI to filter all pirated content would stifle innovation.
3. Policy Focus on Development: The government is promoting AI development. If every AI search product had to compensate for accidental pirated findings, no one would invest in such technology. The court’s ruling signals that as long as platforms are well-intentioned, compliant, and responsive, their innovation will be protected.
---
US Copyright Office’s Viewpoint: RAG as “Copying,” Not Just “Guiding”
In contrast, the US Copyright Office (USCO) took a stricter stance in a 2025 report, differing significantly from Chinese courts:
1. **RAG as “Copying”: USCO views RAG technology as creating a copy of content, regardless of whether it’s just a link. It identifies, retrieves, and displays the content, which constitutes copyright infringement.
2. Fair Use Hard to Apply: Many AI actions are exempt under fair use, but USCO argues that RAG rarely qualifies. Three reasons:
- Lack of Transformation: AI merely reproduces original text or segments without significant processing.
- Excessive Retrieval: AI may retrieve large amounts of content for a single query, exceeding what’s necessary.
- Market Displacement: Users may avoid the original source, affecting the copyright holder’s revenue.
3. Effective Filtering Systems Matter: USCO suggests that if a platform can prove its system effectively filters out pirated content, it may be exempt. For example, Google Books was exempt because it only displayed excerpts and allowed authors to remove them.
In Simple Terms: The USCO believes that providing pirated content, even if unintentional, constitutes infringement unless a platform has a robust filtering system.
---
Revisiting the Mita Case from a US Perspective: Why a Different Outcome?
If the Mita case were judged by USCO, the outcome might be different:
1. From Indirect Liability to Direct Evaluation: Chinese courts consider platforms indirectly responsible, focusing on removal. USCO would evaluate whether the AI’s action itself constituted infringement.
2. Mita’s Filtering System Failed: USCO requires effective filtering. Mita’s system didn’t prevent the pirated link from being displayed, indicating a failure to prevent infringement.
3. Past Actions Could Be Liable: Even if Mita removed the link promptly, its prior actions would still be considered. Since the link was displayed, it was considered a violation of copyright.
---
Global Context: Laws Lagging Behind Technology
This case highlights a broader issue: laws are not keeping up with rapid technological advancements. Neither China nor the US clearly defines whether AI search engines must filter pirated content. Different government departments have varying views:
- Chinese Government: No obligation to filter, as technology isn’t yet capable of doing so.
- US Government: No clear obligation, but failure to filter may lead to infringement liability.
This divergence reflects the complexity of regulating AI in a rapidly evolving industry. Germany’s case shows that responsibilities for AI search are still undefined globally, with varying rulings depending on the country and specific technical details.
---
Implications for Individuals and Businesses:
- For Individuals: Be cautious with AI search results; don’t share pirated content to avoid becoming part of an infringement chain.
- For AI Companies:
- In China, focus on quick response to complaints and effective removal of illegal content.
- Globally, invest in robust filtering systems to prevent and avoid infringement.
- Comply with regulations and maintain proper documentation as a safeguard against legal risks.
---
Conclusion
The Mita case illustrates the core conflict in copyright protection in the AI era: the balance between efficiency of technological innovation and fairness of copyright protection. China prioritizes development, while the US emphasizes strict regulation. Understanding these differences is crucial for both individuals and businesses in the AI industry. As AI becomes more influential, ensuring content security will be a key competitive advantage.