A technology company deploys a large-language model to generate marketing copy, product images, and software documentation at scale. The output reaches millions of Chinese consumers. A competitor reproduces the content without permission. The company moves to enforce its rights – and discovers that the foundational question of who, if anyone, owns that content under Chinese intellectual property legislation has no settled answer.
China's intellectual property legislation has not yet been formally amended to address AI-generated works, leaving ownership, authorship, and enforcement rights to be resolved through court interpretation and emerging regulatory guidance. Chinese courts have begun to recognise that where human creative choices are sufficiently embedded in the generation process, the directing party may hold protectable rights. The doctrinal position remains contested, however, and international businesses operating in or through China face material uncertainty about whether their AI-generated assets are protectable at all.
This analysis examines the doctrinal background, competing judicial interpretations, the gap between statute and practice. Cross-border implications for Asia-Pacific and Middle East clients. Additionally, the strategic steps businesses should consider now – before a dispute crystallises.
Doctrinal background: what Chinese intellectual property law says about authorship
Chinese intellectual property legislation is built on a human-centred model of authorship. The starting premise is that a work must reflect the intellectual creation of a natural person. Corporate entities may hold copyright, but only because a human author's rights have been assigned or because the work was created by an employee within the scope of employment. No statutory provision expressly addresses the situation where the creative output originates primarily from an algorithm rather than a person.
This matters because China's intellectual property legislation sets out categories of protectable works – literary, artistic, musical, cinematographic, photographic, architectural, and others – without addressing computationally generated content. The State Council (the PRC's highest executive body) has issued general policy documents encouraging the development of artificial intelligence, but these instruments do not resolve the authorship question. The SAMR (State Administration for Market Regulation), which oversees certain aspects of digital services and platform conduct, similarly has not issued binding rules on AI-generated content ownership.
Practitioners in China note that the absence of statutory language cuts in two directions. It can be read to exclude AI-generated content from copyright protection entirely. Alternatively, it can be read to leave room for judicial adaptation of existing doctrines – which is exactly what Chinese courts have begun to do. The gap between the two readings is where most commercial risk currently resides.
Patent law presents a related but distinct set of issues. Chinese patent legislation requires an inventor to be a natural person. An AI system cannot be named as inventor. The consequences for international businesses are significant: AI-generated technical solutions may be patentable only if a human inventor can be identified and documented as having made a qualifying inventive contribution. Where the AI's role is predominant, that identification becomes difficult to sustain.
Competing court interpretations and the human-creativity threshold
Chinese courts have delivered a series of decisions that, read together, sketch a provisional and still-contested doctrinal position. No single ruling has been elevated to binding precedent in the way a common law decision would be. Each case turns on its specific facts. Nonetheless, patterns are discernible.
The Beijing Internet Court examined a dispute involving AI-generated imagery and held that copyright may subsist where the plaintiff exercised sufficient human creative choices in selecting parameters, arranging prompts, and editing the output. The court reasoned that the process of directing the AI constituted intellectual creation by a natural person. This approach – sometimes called the "human-in-the-loop" standard – has been adopted in several subsequent first-instance decisions.
A different line of reasoning has emerged from courts that scrutinise the degree of autonomy exercised by the model itself. Where a user simply enters a brief prompt and accepts the output without further creative intervention, some courts have declined to find the human-creativity threshold met. These decisions suggest that the more autonomous the model, the harder it becomes to locate protectable human authorship in the output.
The tension between these two lines is not yet resolved. The China International Court – formally, the China International Commercial Court (CICC) – has jurisdiction over certain cross-border commercial disputes and has signalled openness to considering international developments in AI and intellectual property. However, CICC decisions on this specific question remain limited. The CIETAC (China International Economic and Trade Arbitration Commission) has handled related disputes in the technology licensing context. There. Parties have contractually designated the ownership of AI outputs. a contractual route that offers more certainty than litigation.
Algorithmic accountability adds a further layer of complexity. Where AI output causes harm – for example, by infringing a third party's pre-existing copyright through training-data contamination – questions of software liability arise. Chinese courts have begun to examine whether platform operators owe obligations to police infringing outputs generated by their systems. These obligations sit within the broader body of digital services regulation and are distinct from the copyright ownership question, but they interact with it. A platform that benefits commercially from AI-generated content may face liability for that content under digital services rules even if it cannot itself claim copyright in the output.
For international businesses assessing risk, the practical consequence is this: the human-creativity threshold is a factual question, not a bright-line rule. Documenting the creative choices made by human operators at every stage of the AI-assisted production process is the most reliable way to build a defensible ownership claim under current judicial approaches.
To explore how these court positions intersect with your AI deployment strategy in China, contact us at info@ferrazwhitmore.com.
The gap between statute and practice: where international businesses lose value
The most commercially significant risk arising from doctrinal uncertainty is not the risk of losing a lawsuit. It is the risk of discovering, too late, that valuable assets were never protectable in the first place. A business that invests in AI-generated creative or technical output. without taking steps to establish human-authorship documentation. may find that it cannot enforce rights against copying competitors. Cannot license those assets to third parties on reliable terms. Additionally, cannot include them as protected intellectual property in due diligence for investment or acquisition transactions.
Consider a WFOE (wholly foreign-owned enterprise) operating in China's technology sector. It generates product designs using an AI platform under a technology licensing agreement with a US-based model developer. The licensing agreement assigns all outputs to the WFOE. That contractual assignment is enforceable between the parties. But it cannot create copyright where none exists under Chinese intellectual property legislation. If a Chinese court subsequently holds that the output lacked sufficient human creativity to qualify for protection, the assignment is of limited value against third parties who copy the work.
The gap between contractual allocation and statutory protection is a recurring source of lost opportunity for international businesses. Companies that understand this gap – and structure their workflows accordingly – are positioned to extract full value from AI-generated assets. Those that rely on contractual terms alone, without attending to the human-creativity documentation requirements that Chinese courts now scrutinise, face the prospect of holding unenforceable rights.
A second practical gap concerns enforcement timelines. Even where a court recognises protectable rights in an AI-generated work, enforcement in China is a multi-stage process. Administrative channels through the SAMR or local intellectual property offices can move faster than civil litigation. Courts with specialised intellectual property jurisdiction – the Beijing, Shanghai, and Guangzhou intellectual property courts – offer expertise but are burdened with high caseloads. First-instance proceedings may conclude within six to eighteen months. Appeals extend that timeline. For businesses in fast-moving technology sectors, that timeline may exceed the commercial life of the contested asset.
A third gap concerns the training-data question. AI models trained on third-party works may generate output that reproduces protectable expression from those source works. Chinese intellectual property legislation does not yet contain a specific exception for AI training data analogous to the text-and-data mining exceptions found in some other jurisdictions. The absence of such an exception means that the legal basis for training on commercially published content without a licence remains uncertain. Businesses that develop or deploy AI systems in China should conduct a training-data audit as part of their AI Act compliance and risk-management processes. even where the applicable AI regulation originates outside China. because infringement claims can be brought in Chinese courts by Chinese rights holders whose works were used in training.
For a detailed comparison of how AI-generated content ownership is treated in another high-growth market. See our analysis of AI-generated works and intellectual property in the UAE. There, a distinct regulatory approach has emerged under Gulf commercial legislation.
Cross-border implications for Asia-Pacific and Middle East clients
For businesses operating across multiple jurisdictions in Asia and the Middle East, China's emerging position on AI-generated works creates several layers of cross-border complexity.
The first concerns the interaction between Chinese intellectual property rules and the laws of the jurisdiction where a business is headquartered or where its AI outputs are primarily exploited. A Singapore-based company using a Chinese AI platform to generate content for sale in multiple markets cannot assume that copyright recognised in Singapore will translate into enforceable rights in China. Each jurisdiction applies its own authorship rules. Where the human-creativity threshold is met in one jurisdiction but not another, the business holds geographically fragmented rights – commercially valuable in some markets, unenforceable in others.
The second concerns technology licensing structures. Many international businesses access Chinese AI tools through licensing agreements governed by Chinese commercial legislation. The allocation of AI output rights in those agreements must be drafted with the current judicial landscape in mind. A clause that simply assigns "all outputs" to the licensee does not resolve whether those outputs are protectable under Chinese intellectual property legislation. Bespoke language addressing the human-creativity documentation obligation – and specifying who bears the burden of creating and maintaining that documentation – is increasingly necessary.
The third concerns dispute resolution. Where a cross-border AI-related intellectual property dispute arises, the choice of forum matters significantly. CIETAC arbitration offers confidentiality and a degree of procedural predictability for international parties. Chinese courts, including the CICC, offer state enforcement power but apply Chinese procedural rules. Parties who fail to include an effective dispute-resolution clause in their AI-related agreements may find themselves litigating in a forum they did not choose, under rules that are still developing.
The fourth concerns the interaction with AI regulation developing elsewhere. The EU's AI Act, while not directly applicable in China, creates compliance obligations for businesses that deploy AI systems in EU markets. Where those same businesses operate in China, they face a dual-regime challenge: meeting the AI Act's requirements for algorithmic accountability and transparency while also managing the distinct requirements of Chinese AI regulation. The State Council has issued interim measures for generative AI services, and these measures impose obligations on operators of generative AI platforms directed at Chinese users. Compliance with one regime does not guarantee compliance with the other.
A fifth dimension concerns market entry. Foreign businesses entering the Chinese AI market through a WFOE structure must consider how their intellectual property assets – including AI-generated content – will be protected from the outset. The SAMR registration process for technology-related businesses now requires disclosure of AI tools used in the business's operations in certain sectors. This disclosure obligation creates a paper trail that can later become relevant in intellectual property disputes. Businesses that manage this disclosure carefully, and that align it with their human-creativity documentation practices, are better positioned to assert rights in their AI-generated assets.
For a comprehensive view of how our AI and technology law practice in China can support your cross-border AI strategy, reach out to our team at info@ferrazwhitmore.com.
Strategic recommendations and the regulatory outlook
The doctrinal uncertainty that currently characterises AI-generated works in China is unlikely to persist indefinitely. The State Council has signalled that a review of intellectual property legislation to address AI-related questions is under consideration. Legislative amendments, when they come. May adopt any of several approaches: recognising a new category of computer-generated works without a human authorship requirement. codifying the human-in-the-loop standard that courts have begun to apply. or declining to extend copyright to AI outputs while creating a separate sui generis protection. Each outcome would produce different commercial consequences for businesses that have already built AI-generated content into their product and service offerings.
In the meantime, the following strategic steps reflect current best practice for international businesses with material AI-generated IP exposure in China.
Document human creative choices at every stage. Maintain records showing the prompts used, the parameters set, the editorial choices made in selecting among multiple AI outputs, and any post-generation modifications. This documentation is the foundation of any future copyright claim before a Chinese court.
Audit training data before deployment. Where your business operates or plans to operate an AI system that was trained on third-party content. Conduct a training-data review to identify potential infringement exposure under Chinese intellectual property legislation. The absence of a specific text-and-data mining exception in Chinese law means that this exposure is real and currently unquantified for many businesses.
Review technology licensing agreements for output-allocation language. Standard-form technology licensing agreements drafted outside China frequently contain generic output-assignment clauses that do not address the human-creativity threshold. These agreements should be reviewed and, where necessary, amended to include documentation obligations and dispute-resolution provisions suited to the Chinese legal context.
Consider CIETAC arbitration clauses for AI-related commercial agreements. For businesses entering into AI-related commercial relationships in China. Specifying CIETAC as the dispute-resolution forum. with a governing law clause aligned to the nature of the agreement. provides procedural predictability and enforceability that ad hoc litigation does not.
Monitor the State Council's legislative programme. Proposed amendments to Chinese intellectual property legislation affecting AI-generated works could move quickly once introduced. Businesses should track the legislative programme and assess how proposed changes would affect the protectability of existing AI-generated assets.
Align Chinese IP strategy with broader AI Act compliance obligations. For businesses subject to the EU's AI Act or to comparable regulatory regimes in other jurisdictions. Aligning the algorithmic accountability and documentation requirements of those regimes with the human-creativity documentation requirements of Chinese courts creates operational efficiency. A single documentation standard, developed for the most demanding applicable regime, is often the most practical approach.
The businesses most exposed to lost opportunity in this area are those that treat AI-generated content as automatically owned and automatically protectable. The businesses best positioned to capture value are those that have structured their AI workflows to produce the evidence that Chinese courts currently require.
For a preliminary review of your AI-generated intellectual property position in China, email info@ferrazwhitmore.com.
Frequently asked questions
Q: Can a company own copyright in an AI-generated work under Chinese intellectual property legislation?
A: Chinese intellectual property legislation currently requires a human author for copyright to subsist. Courts have, however, recognised that a company directing the AI tool may hold rights in the output where substantial human creative choices shaped the result. The question of ownership turns on the degree of human involvement documented during the creation process.
Q: How long does it take for Chinese courts to resolve an AI copyright dispute, and what does it cost?
A: First-instance proceedings before an internet court or specialised intellectual property tribunal in China typically conclude within six to eighteen months. Costs depend on claim complexity, but legal fees for commercial IP disputes routinely reach tens of thousands of US dollars at first instance, with appeals adding further time and expense.
Q: Is it a misconception that AI-generated content automatically belongs to the AI developer in China?
A: Yes. Ownership does not default automatically to the developer of the underlying model. Chinese courts have examined the contractual terms of the technology licensing agreement between developer and user, the nature of user inputs, and the degree of human editorial control. The developer, the platform operator, or the end user may each hold a colourable claim depending on the facts.
About Ferraz & Whitmore
Ferraz & Whitmore is an international law firm based in Lisbon, advising business clients across 46 jurisdictions. Our team combines Portuguese civil law expertise with English common law tradition to deliver cross-border legal solutions in AI and technology law, intellectual property protection, and digital services regulation. In China and across Asia-Pacific markets, we advise international businesses – including WFOEs, technology licensors, and institutional investors – on AI-generated content ownership, software liability, algorithmic accountability, and cross-border IP enforcement. Our AI law practice covers both the doctrinal questions that arise under Chinese intellectual property legislation and the practical compliance obligations imposed by the State Council's generative AI measures and comparable regimes in other jurisdictions. Engaging a lawyer in China with cross-border experience across civil law and common law systems is particularly valuable where AI-generated asset protection must be maintained simultaneously in multiple markets. As an international law firm advising on China matters, Ferraz & Whitmore supports clients from the initial IP audit through to dispute resolution before CIETAC and the Chinese courts. Our attorneys have advised on intellectual property matters in China spanning copyright, patent, and trade secret disputes across both domestic and cross-border contexts. To discuss your AI-generated IP position in China, contact us at info@ferrazwhitmore.com.
Disclaimer: This publication is provided for informational purposes only and does not constitute legal advice. The information herein should not be relied upon as a substitute for professional legal counsel tailored to your specific circumstances. Ferraz & Whitmore assumes no liability for actions taken or not taken based on the contents of this material. For advice regarding your particular situation, please contact info@ferrazwhitmore.com.