A technology company based in Singapore deploys a generative AI system to produce thousands of marketing images, product descriptions, and software components for distribution across Asian markets. Its Hong Kong subsidiary registers the outputs with the Companies Registry Hong Kong as business assets, licenses them to regional partners, and pursues infringers through the Hong Kong High Court. Then a dispute arises. A licensee refuses to pay, arguing that AI-generated outputs cannot be owned under Hong Kong intellectual property legislation. The company's lawyers discover that no court has yet resolved the question. Every commercial assumption underpinning the deal suddenly looks fragile.
AI-generated works occupy a genuinely uncertain position under Hong Kong intellectual property legislation. The existing body of law contains a specific provision for computer-generated works. attributing authorship to the person who makes the necessary arrangements for creation. but that provision was drafted for earlier generations of software. Not for autonomous generative AI. The Hong Kong High Court has not yet issued a definitive ruling on whether modern AI outputs qualify for protection, and no legislative reform has been enacted to close the gap.
This analysis examines the doctrinal foundations, the competing interpretations that practitioners and businesses are navigating, the gap between statute and current practice. The cross-border dimensions for clients operating across Asia and the Middle East. Additionally, the strategic steps that international businesses should take now. before the law is settled.
Doctrinal foundations: what the existing legislation actually says
Hong Kong intellectual property legislation follows the common law tradition inherited from English law. Copyright protection attaches to original works. Originality, in the common law sense, requires that a work originates from an author and reflects some degree of skill, judgment, or labour. This requirement is entirely author-centric. It assumes a human being at the source of creation.
The legislation does, however, contain a notable departure from this baseline. It provides that, for a computer-generated work, the author is taken to be the person who makes the necessary arrangements for the work to be created. This provision was introduced to address situations where a computer assists in the creative process. for example. There. A programmer writes code that generates a fractal image. Alternatively. There, a journalist uses software to compile a statistical chart. The human who arranged for the computer to produce the output was deemed the author.
Two further features of this provision matter commercially. First, the protection term for computer-generated works is significantly shorter than for works with a human author. It runs from the date of creation. Second, the provision is silent on what degree of human input is required to qualify as "making the necessary arrangements." This silence is now the central doctrinal fault line for AI-generated content.
Practitioners in Hong Kong note that the provision was never tested against a system capable of producing substantive creative output with minimal human instruction. A prompt of ten words that generates a 5,000-word article or a photorealistic image raises qualitatively different questions from a programmer who writes thousands of lines of code to produce a specific visual output. The gap between the statutory text and current technological reality is substantial.
Beyond copyright, Hong Kong intellectual property legislation also covers trade marks, patents, and design rights. Each of these regimes presents its own challenges for AI-generated subject matter. Patent law, for example, requires an inventor. and the question of whether an AI system can be named as inventor. Alternatively. Whether the human operator must be treated as the inventor, has produced divergent positions across common law jurisdictions. The Hong Kong position has not yet been authoritatively resolved, but the prevailing view among specialists is that human inventorship remains required under current legislation.
Competing interpretations and the gap between statute and practice
Three distinct interpretive positions have emerged in Hong Kong legal practice. Each carries different implications for businesses building commercial strategies on AI-generated content.
The first position treats the computer-generated works provision broadly. On this view, whoever operates the AI system – selecting the model, constructing the prompts, curating the outputs – is making the "necessary arrangements." This person is the author. Protection attaches. The argument draws support from the original legislative purpose, which was to extend protection rather than deny it. It also aligns with the commercial expectations of businesses that invest in AI infrastructure and treat the outputs as proprietary assets.
The second position reads the provision narrowly. It argues that "necessary arrangements" requires meaningful creative contribution, not merely the act of pressing a button or entering a generic prompt. On this view, a one-line prompt that generates a complex image does not suffice. The human input must be substantial enough that the resulting work reflects the human's own intellectual effort. This interpretation draws on the originality doctrine and on the underlying policy concern that copyright should not be a windfall for minimal human effort.
The third position – increasingly cited by practitioners focused on algorithmic accountability – argues that fully autonomous AI outputs fall outside the existing provision entirely. The provision, on this view, was drafted for deterministic software: tools that reliably produce a predictable output from a given input. Generative AI systems are probabilistic. They produce different outputs from identical inputs. The "person making the necessary arrangements" cannot be said to have arranged for any specific work, because the specific work was not foreseeable. This reading would leave fully autonomous AI outputs without copyright protection in Hong Kong.
The Hong Kong High Court has not yet been called upon to choose between these positions in a case directly addressing generative AI. Courts in comparable common law jurisdictions – including England and Australia – have begun to address related questions, and their reasoning is persuasive authority in Hong Kong. The English courts have signalled a preference for reading the computer-generated works provision purposively, but have also emphasised that the degree of human creative input is relevant to whether any protection arises. Hong Kong courts are likely to follow a similar path.
In practice, this uncertainty has produced a bifurcated market response. Larger technology companies and institutional content producers take the broad view and proceed as if their AI-generated outputs are protected. They register trade marks for AI-created logos, license AI-generated text and imagery on commercial terms, and pursue infringers. Smaller operators and in-house legal teams are more cautious, sometimes declining to rely on AI-generated works as primary IP assets and instead retaining human-authored versions as backup.
Software liability questions compound the picture. When an AI system generates content that infringes a third party's copyright – for example, by reproducing substantial elements of a training dataset – the question of who bears liability is unsettled. The AI developer, the operator, and the end user may each face exposure under different theories. Technology licensing agreements in the Hong Kong market are increasingly addressing this risk through indemnity clauses and output audit obligations, but standard-form agreements often leave the allocation unclear.
For a tailored strategy on AI-generated intellectual property protection in Hong Kong, reach out to our AI and technology law practice in Hong Kong for a preliminary assessment of your situation.
Training data, infringement exposure, and digital services
A dimension that has received increasing attention in Hong Kong is the copyright status of training data. Generative AI systems are trained on large corpora of existing works. In many cases, those works are protected by copyright. The act of ingesting, copying, and processing them for training purposes may itself constitute infringement under Hong Kong intellectual property legislation, unless a relevant exception applies.
Hong Kong's copyright exceptions include provisions for research, education, and – significantly – text and data mining. However, the text and data mining exception in Hong Kong is narrower than analogous provisions in some other jurisdictions. It applies primarily to non-commercial research. A technology company training a commercial AI model on scraped web content does not automatically benefit from this exception. The commercial character of the training activity is a material disqualifying factor.
This creates a significant exposure for businesses that have built AI systems using broad training datasets without clearing rights. If the training process infringed third-party copyright, the resulting AI model may itself be a vehicle for ongoing infringement. Every output the model generates that reproduces a substantial part of a training work could constitute a separate infringing act. The Hong Kong High Court has not yet addressed this chain-of-infringement theory in the AI context. However. The doctrinal foundation for such a claim exists in the general principles of secondary liability under Hong Kong intellectual property legislation.
Digital services businesses operating in Hong Kong – including platforms that host and distribute AI-generated content – face related exposure as intermediaries. Platform liability under Hong Kong intellectual property legislation depends on whether the platform has actual or constructive knowledge of infringing material and whether it took expeditious steps to remove it. AI-generated content complicates this calculus. Platforms cannot always identify whether AI-generated material reproduces protected works. Standard notice-and-takedown procedures are not well-adapted to the probabilistic nature of AI output.
The Securities and Futures Commission (SFC) has separately flagged concerns about AI-generated content in financial services contexts – particularly AI-generated investment research, marketing materials, and client communications. While the SFC's primary concern is accuracy and disclosure rather than intellectual property, the regulatory attention signals that AI-generated outputs in regulated industries attract scrutiny from multiple directions simultaneously. A business that deploys AI across both creative and financial services contexts must manage both IP and regulatory dimensions in parallel.
Practitioners handling technology licensing transactions in Hong Kong are now routinely including provisions that address: the ownership of AI-generated outputs. the permitted use of licensee data for AI training. indemnification for third-party IP claims arising from AI outputs. and audit rights over AI systems used in connection with licensed technology. These clauses are increasingly non-negotiable in transactions involving significant AI deployment.
Cross-border dimensions for Asia-Pacific and Middle East clients
For international businesses, the Hong Kong IP position does not exist in isolation. Hong Kong is a gateway jurisdiction for mainland China, Southeast Asia, and increasingly for Middle Eastern investors structuring regional IP holding arrangements. The interaction between Hong Kong law and the IP regimes of adjacent jurisdictions creates both opportunities and risks.
Mainland China has taken a more assertive approach to AI-generated works than Hong Kong. Chinese courts have issued rulings – the first of their kind in Asia – recognising copyright in certain AI-assisted works where the human operator exercised sufficient selection and arrangement. This divergence means that a business that cannot establish copyright in Hong Kong may nonetheless have protectable rights in China for the same work, provided the human creative contribution meets the Chinese standard. Cross-border IP strategies for the Greater Bay Area must account for this divergence explicitly.
Singapore has similarly moved ahead of Hong Kong in issuing guidance on AI and intellectual property. The Singaporean approach emphasises human authorship as a baseline requirement but acknowledges that AI-assisted works, where the human makes creative choices, can qualify for protection. Businesses operating across Hong Kong and Singapore should map their AI workflows carefully to identify where protectable human contribution occurs and document it contemporaneously.
For clients with Middle Eastern operations – particularly those structuring IP holding companies in jurisdictions with developing AI regulation – Hong Kong's uncertainty has a specific implication. If an IP holding arrangement depends on Hong Kong copyright subsisting in AI-generated works, and that copyright is later found not to exist, the commercial rationale for the holding structure collapses. Transfer pricing arrangements and royalty streams built on that assumption may also be challenged by tax authorities. The interaction between IP law uncertainty and tax structuring is a risk that is frequently underestimated in regional expansion planning.
Arbitration of IP disputes in Hong Kong typically proceeds before the Hong Kong International Arbitration Centre (HKIAC). HKIAC has developed considerable expertise in technology and IP disputes. International businesses involved in cross-border licensing of AI-generated content should consider whether HKIAC arbitration clauses offer advantages over litigation in the Hong Kong courts. particularly in terms of confidentiality. Enforceability of awards across the region. Additionally, the ability to appoint arbitrators with specialist AI expertise. For a comparative perspective on how AI-generated IP questions are being addressed in other major hubs, the analysis of AI-generated works and intellectual property in the UAE offers a useful cross-jurisdictional reference point.
A multinational technology group headquartered outside Hong Kong that licenses AI-generated creative works to Hong Kong distributors should also consider the enforceability of those licences if the underlying copyright is challenged. A licence to exploit a work that turns out not to be protected is an agreement without subject matter. The licensee's obligation to pay royalties evaporates. Representations and warranties in the licensing agreement – particularly those relating to the validity and subsistence of the licensed rights – become the primary line of protection for the licensor.
To discuss how Hong Kong intellectual property law and AI regulation apply to your cross-border licensing strategy, contact us at info@ferrazwhitmore.com.
Strategic recommendations and the regulatory outlook
The unsettled state of Hong Kong law is not a reason for commercial paralysis. It is, however, a reason for deliberate risk management. The following considerations should inform how businesses structure their AI-IP strategies in Hong Kong now.
First, document the human creative process at every stage of AI-assisted production. Where a human selects, arranges, refines, or curates AI outputs, that process should be recorded. Contemporaneous records of human editorial decisions are the most defensible basis for asserting copyright under either the broad or narrow reading of the computer-generated works provision. Businesses that cannot demonstrate any human creative contribution are exposed under every interpretive position.
Second, conduct a training data audit before commercialising AI models in Hong Kong. The copyright status of training datasets is not merely a historical question. It affects the ongoing legality of operating and licensing the model. A training data audit identifies which elements of the corpus carry third-party rights, assesses whether applicable exceptions cover their use. Additionally. Allows businesses to manage or mitigate residual exposure through licensing, model modification, or selective deployment.
Third, review technology licensing agreements to ensure they address AI-specific risks explicitly. Standard software licence terms – drafted before generative AI became mainstream – typically do not allocate IP ownership of AI-generated outputs, address training data obligations, or apportion liability for AI-generated infringement. Bespoke provisions are now commercially standard in sophisticated transactions. Relying on legacy agreements for AI deployments creates gaps that counterparties can exploit.
Fourth, consider defensive registration strategies. Trade mark registration in Hong Kong protects logos, brand elements, and other identifiers regardless of whether the underlying design was human- or AI-generated. Where AI-generated brand assets can be registered, registration provides a layer of protection that does not depend on resolving the copyright question. Design registration may offer a similar fallback for three-dimensional AI-generated product designs.
Fifth, monitor legislative developments closely. The Hong Kong government has signalled interest in reviewing the intellectual property legislative regime in the context of AI. Any reform to the computer-generated works provision – whether to extend, restrict, or clarify it – will materially affect the commercial value of AI-generated IP assets in Hong Kong. Businesses with significant exposure should participate in consultation processes and adjust their strategies promptly when reform is enacted.
The broader regulatory trajectory points toward greater, not lesser, scrutiny of AI systems. The international momentum around AI Act compliance. driven primarily by the European Union but increasingly influential on the thinking of regulators in Asia. is reshaping how businesses document, audit, and disclose their AI systems globally. Hong Kong regulators have not enacted AI-specific legislation, but the SFC, the Hong Kong Monetary Authority, and other sector regulators have issued guidance on AI use that emphasises transparency, accountability, and risk management. Algorithmic accountability is becoming an expected standard of practice in regulated industries, and that expectation is migrating into the IP context as courts and regulators ask how AI systems produce the outputs they do.
The intersection of AI Act compliance considerations, technology licensing obligations, and software liability exposure means that businesses cannot manage AI-related IP risk in isolation. A coordinated legal strategy – covering copyright, trade marks, data protection, licensing, and regulatory compliance – is the only approach that reliably protects commercial value in AI-generated works as the law develops. For a comprehensive view of the intellectual property protections available in Hong Kong beyond the AI context. The full range of intellectual property services for Hong Kong provides the broader framework within which AI-specific IP strategy sits.
Frequently asked questions
Q: Does Hong Kong copyright law protect works generated entirely by an AI system?
A: Hong Kong intellectual property legislation contains a provision for computer-generated works that may extend protection even where no human author is identifiable, attributing authorship to the person who makes the necessary arrangements. However, courts have not yet ruled definitively on whether this provision covers outputs from autonomous AI systems. Until the Hong Kong High Court or the legislature clarifies the position, the scope of protection for fully autonomous AI outputs remains genuinely uncertain.
Q: How long does copyright protection last for a computer-generated work in Hong Kong?
A: Under Hong Kong intellectual property legislation, computer-generated works attract a shorter protection period than works with a human author. The term runs from the date of creation rather than from the death of an author. Businesses relying on this provision should factor this shorter window into their IP asset planning and technology licensing strategies.
Q: Who owns the IP rights in an AI-generated work produced by an employee using a company-licensed AI tool?
A: Ownership depends on the interaction between employment legislation, the technology licensing terms of the AI tool, and any work-made-for-hire provisions in the employment contract. In many cases the employer will have the stronger claim, but AI tool vendors increasingly assert licence-level restrictions on commercial use of outputs. A thorough review of both the employment agreement and the software licence is essential before asserting ownership or exploiting the work commercially.
About Ferraz & Whitmore
Ferraz & Whitmore is an international law firm based in Lisbon, advising business clients across 46 jurisdictions. Our Asia-Pacific practice supports technology companies, institutional investors, and in-house legal teams on AI and technology law, intellectual property strategy, and cross-border digital services in Hong Kong and across the region. We combine Portuguese civil law expertise with English common law tradition to deliver integrated advice on AI-generated works, technology licensing, software liability, and algorithmic accountability in markets where these questions are commercially live. As an international law firm with established experience in Hong Kong, we advise clients who need a lawyer in Hong Kong with cross-border capability – not just knowledge of a single legal system. The firm's AI and technology law practice covers regulatory compliance, IP structuring, and dispute strategy across both civil and common law systems. To explore how we can support your AI-IP strategy in Hong Kong, 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.