A technology company based in Southeast Asia deploys a generative AI system to produce marketing copy, product images, and software documentation at scale. The outputs reach the market. A competitor reproduces several of those outputs verbatim. When the company's legal team investigates, it confronts a question that Singapore's intellectual property legislation has not yet answered with certainty: does copyright even subsist in works produced without direct human authorship? The commercial stakes are real. The legal answer is not yet settled.
Singapore's intellectual property legislation anchors copyright protection in the concept of a human author. Where an AI system generates a work autonomously, the subsistence of copyright – and the identity of any owner – remains doctrinally unresolved under current statute and case law. Businesses operating in Singapore should treat AI-generated outputs as assets of uncertain legal status until ownership is secured by contract or the legislature clarifies the position.
This analysis examines the doctrinal foundations, the gap between statute and practice, the emerging positions of the Singapore High Court and specialist arbitral bodies. Cross-border implications for Asia-Pacific and Middle Eastern clients. Additionally, the strategic steps that businesses can take now to protect value while the law develops.
Doctrinal foundations: authorship, originality, and the human-author requirement
Singapore's intellectual property legislation, including its copyright legislation, was substantially revised in the years leading up to 2022. The revised regime modernised several provisions. It did not, however, introduce a dedicated rule for AI-generated works.
The concept of authorship in Singapore copyright law is rooted in common law tradition. Protection subsists in original works made by authors. An author, in the common law sense, is a human being – or a legal entity to which authorship rights are attributed by statute or agreement. An AI system is neither. It cannot hold rights, bear obligations, or be identified as a legal person under Singapore's Companies Act (Singapore corporate legislation) or any analogous instrument.
Originality in Singapore copyright law follows the "independent creation" standard rather than the higher "creative originality" threshold applied in some civil law systems. The work must originate from its author, meaning it must not be copied and must involve at least a minimal degree of intellectual effort. Where the work is generated entirely by an algorithm, it is genuinely contested whether that standard is satisfied at all. The algorithm does not "create" in the legal sense. It applies statistical weights to training data.
Some practitioners argue that a sufficiently detailed human prompt constitutes the intellectual effort required for authorship. The prompt-writer, on this view, is the author because the expressive choices are encoded in the instructions given to the system. Others counter that authorship requires control over the form of expression, not merely the specification of inputs. On this second view, the prompt-writer is more analogous to a client who commissions a painting than to the artist who holds the brush.
The Singapore courts have not yet issued a definitive ruling on which interpretation prevails. The Singapore High Court has addressed authorship and originality in the context of computer-assisted works – works where a human uses software as a tool. Those decisions suggest that courts look to the human whose intellectual effort is most directly expressed in the output. Applying that reasoning to fully autonomous AI outputs is a logical extension, but it has not been confirmed.
A further doctrinal complexity arises from the work-made-for-hire and commissioning rules. Under Singapore's intellectual property regime, copyright in certain commissioned works vests in the commissioning party rather than the creator. If that rule is extended by analogy to AI outputs, the entity that procured the AI service might hold any copyright that subsists. This reasoning is commercially attractive but legally untested. Businesses seeking to rely on it should take specialist advice on intellectual property in Singapore before assuming ownership.
The training-data problem: infringement, fair dealing, and algorithmic accountability
Before a generative AI system produces any output, it must be trained. Training at commercial scale involves ingesting vast quantities of existing content – text, images, audio, code. Much of that content is protected by copyright. Reproducing it without permission may constitute infringement under Singapore's intellectual property legislation, subject to specific exceptions.
Singapore's copyright legislation provides a fair-dealing defence for purposes such as research, criticism, and reporting. It does not provide a broad text-and-data-mining exception. Several jurisdictions – notably the European Union and Japan – have introduced specific carve-outs permitting AI training on lawfully accessed data. Singapore has not followed suit as of the date of this analysis.
This gap creates exposure for businesses that train models on scraped or licensed datasets without granular rights analysis. The risk is not merely theoretical. Rights-holders in the creative industries and software sectors have commenced litigation in multiple common law jurisdictions challenging training-data practices. Singapore-incorporated entities running training operations could face similar claims before the Singapore High Court or in arbitration proceedings under the Singapore International Arbitration Centre (SIAC) rules, where IP-related disputes are increasingly submitted.
Algorithmic accountability adds a further dimension. Where an AI system generates output that closely resembles a protected work – because the training data was insufficiently filtered – the question of software liability arises. The developer, the operator, and the end-user may each face exposure depending on how liability is allocated in their technology licensing agreements. Practitioners in Singapore note that standard software licensing terms drafted before the generative AI era frequently contain no provisions addressing this risk. Revisiting those agreements is an immediate practical step.
The Monetary Authority of Singapore (MAS) has published guidance on responsible AI use in the financial sector. That guidance addresses bias, transparency, and explainability. It does not resolve copyright questions directly, but it reflects a regulatory posture in which algorithmic accountability is treated as a distinct governance obligation – separate from and additional to intellectual property compliance.
For businesses registered with the Accounting and Corporate Regulatory Authority (ACRA), the corporate governance implications of AI deployment are also coming into focus. Directors who authorise AI-driven production processes without adequate IP clearance may face scrutiny under their fiduciary duties if those processes later generate material liability. This is an area where technology law and corporate law converge in ways that many boards have not yet addressed.
Ownership disputes and strategic gap between statute and practice
Even if copyright subsists in an AI-generated work, ownership is not self-evident. At least three parties may advance a credible claim: the AI developer, the user who crafted the prompts, and the business entity on whose behalf the system was deployed. Where these parties are in different jurisdictions, the ownership question intersects with private international law rules.
Singapore's conflict-of-laws doctrine for intellectual property generally applies the law of the jurisdiction where protection is sought – the lex loci protectionis (the law of the place of protection). For Singapore copyright, that means Singapore law governs. But if the developer is incorporated in the United States and the prompt-writer is an employee of a Japanese company, the contract between them may be governed by a third system. Practitioners in Singapore advise that multi-party AI deployment agreements should include express choice-of-law clauses and explicit intellectual property assignment provisions covering AI outputs.
The gap between statute and practice is most acute in the absence of express assignment. Singapore's intellectual property legislation does not address AI authorship directly. Courts may fill the gap by analogy with existing rules. They may alternatively hold that no copyright subsists at all – leaving AI outputs in the public domain. Neither outcome is commercially satisfactory for businesses that have invested in AI-driven content production.
Several practical scenarios illustrate the risk. A media company uses a generative AI system to produce thousands of news summaries. A competitor reproduces those summaries. Without copyright, the media company has no infringement claim. Its investment in the AI system – and the commercial value of the output – is legally unprotected. Alternatively, a software developer uses an AI tool to generate code. The tool is licensed under terms that assign all outputs to the licensor. The developer deploys the code in a client project. The client later discovers that it does not own the software it paid for.
Technology licensing agreements are therefore the primary risk-management instrument at present. A well-drafted license will address: ownership of outputs, warranties regarding training-data rights, indemnities for third-party IP claims, and audit rights. These provisions are standard in mature software agreements but are often absent or ambiguous in AI-specific agreements because the market moved faster than legal practice.
For businesses with operations across Asia and the Middle East, a comparison with the UAE's approach is instructive. The UAE has taken steps toward recognising AI-generated works in specific regulatory contexts, though its intellectual property legislation is also in transition. A detailed comparative analysis is available in our deep analysis of AI-generated works and intellectual property in the UAE.
To explore how technology licensing and IP ownership structures can be tailored to your AI deployment in Singapore, contact us at info@ferrazwhitmore.com.
Cross-border implications for Asia-Pacific and Middle Eastern clients
Singapore functions as a regional hub for technology businesses operating across Southeast Asia, South Asia, and the broader Asia-Pacific region. Many of those businesses structure their intellectual property holding entities in Singapore specifically to benefit from its bilateral tax treaty network and its membership of international intellectual property conventions. The uncertainty around AI-generated works therefore has implications well beyond Singapore's territorial borders.
Under the Berne Convention (international copyright treaty framework), copyright protection in one member state is extended to nationals of other member states on the basis of national treatment. If Singapore's courts ultimately hold that AI-generated works do not attract copyright, that holding may affect whether protection is available in other Berne member states where the work is exploited. The converse is also true: a jurisdiction that extends protection to AI-generated works may provide rights that Singapore law denies.
For businesses holding IP assets in Singapore for regional exploitation, this creates a portfolio management challenge. An AI-generated work that is unprotected in Singapore but protected in another jurisdiction requires a jurisdiction-specific registration and enforcement strategy. The cost of maintaining that strategy across multiple jurisdictions is significant. It strengthens the commercial case for ensuring that human authorship – in the form of substantive creative intervention – is built into AI-assisted production workflows wherever commercially feasible.
SIAC arbitration is a frequent choice for resolving technology and IP disputes across the Asia-Pacific region. SIAC's procedural rules permit consolidation of related disputes and emergency arbitration. For disputes involving AI-generated works, the doctrinal uncertainty at the substantive level means that arbitrators may have considerable interpretive latitude. Parties to AI deployment agreements should consider including express substantive provisions addressing AI output ownership in their arbitration clauses, rather than leaving those questions for an arbitral tribunal to resolve without contractual guidance.
MAS regulation of digital services in the financial sector adds a further cross-border dimension. Financial institutions using generative AI to produce investment research, customer communications, or compliance documentation face both IP ownership questions and MAS conduct-of-business obligations. Where AI-generated financial content is reproduced or redistributed across jurisdictions, both sets of rules apply simultaneously. A regulatory strategy that addresses only one dimension while neglecting the other is incomplete.
For clients based in the Gulf Cooperation Council region who use Singapore-domiciled AI platforms to produce Arabic-language content for regional distribution. The interaction between Singapore IP law, UAE copyright law. Additionally, Saudi Arabian intellectual property legislation creates a genuinely complex multi-layer structure. Rights clearance and ownership documentation need to be addressed in all three systems, not just the jurisdiction where the AI platform is incorporated.
AI Act compliance is also entering the conversation for Singapore businesses with European operations or European customers. The EU's AI regulation imposes transparency, documentation, and risk-classification obligations on AI systems deployed in EU-facing contexts. Those obligations do not themselves resolve IP ownership, but they require detailed records of training data provenance and model governance – records that are also directly useful in IP disputes. Building AI Act compliance infrastructure can therefore serve double duty as IP risk management.
Strategic recommendations and the regulatory outlook
The doctrinal uncertainty described in this analysis is not a permanent condition. Singapore's legislature has shown willingness to respond to technological change with targeted statutory reform. The 2021 copyright reforms are evidence of that responsiveness. A further round of reform addressing AI-generated works is widely anticipated by practitioners in Singapore, though no firm timeline has been confirmed.
Several strategic steps are available to businesses operating in the interim period.
First, build human authorship into AI-assisted workflows wherever the commercial value of the output justifies the investment. This means ensuring that a human author makes substantive creative choices that are expressed in the final work – not merely specifying a prompt and accepting the output unchanged. Human editorial review that results in material modification of AI-generated content strengthens the authorship claim considerably.
Second, address IP ownership expressly in every technology licensing agreement and every AI deployment contract. The agreement should specify who owns outputs, whether training data was cleared, what warranties the licensor provides against third-party IP claims, and what happens if a court subsequently holds that no copyright subsists.
Third, conduct a training-data rights audit for any AI system deployed at commercial scale within the business. This audit should identify the sources of training data, the terms under which it was accessed, and whether those terms permit use for machine-learning purposes. Where gaps are identified, remediation may involve acquiring additional licenses, switching to a system trained on licensed or public-domain data, or accepting and documenting a reasoned fair-dealing position.
Fourth, register designs and trade marks for AI-generated visual elements wherever possible. These rights do not depend on the authorship concept in the same way as copyright. A distinctive AI-generated logo may be registrable as a trade mark even if its copyright status is uncertain. Design registration provides an additional layer of protection for product aesthetics.
Fifth, engage with the ongoing public consultation processes. Singapore's Intellectual Property Office has been an active participant in international discussions on AI and intellectual property. Businesses with significant AI-driven content operations have an interest in contributing to those discussions to shape the outcome of future reforms.
The regulatory outlook beyond Singapore also warrants monitoring. The WIPO process on AI and intellectual property has produced substantial documentation of divergent national positions. International consensus on a minimum standard for AI-generated works – if it emerges – would significantly reduce the cross-border complexity currently faced by businesses operating across multiple legal systems in the Asia-Pacific region.
For businesses that move first to establish clear contractual and governance structures, the current period of uncertainty is an opportunity rather than a purely defensive challenge. Competitors who delay will face greater disruption when the law is clarified. Well-structured IP portfolios built now – even in conditions of legal uncertainty – will be more resilient and more valuable when that clarity arrives.
To discuss a tailored IP and AI governance strategy for your operations in Singapore, reach out to info@ferrazwhitmore.com.
Self-assessment: when these questions apply to your business
The issues addressed in this analysis are directly relevant if one or more of the following conditions apply to your business.
- Your business generates content, code, or design assets using AI systems at commercial scale in or through Singapore.
- You hold intellectual property assets in Singapore for regional exploitation across Asia-Pacific or the Middle East.
- Your technology licensing agreements predate the widespread adoption of generative AI and have not been reviewed for AI-specific provisions.
- Your AI training data includes third-party materials whose copyright status has not been assessed against Singapore's fair-dealing rules.
- You operate in a regulated sector – financial services, healthcare, or media – where MAS or sector-specific guidance imposes AI governance obligations alongside IP considerations.
Before taking a position on any of the issues above, verify the following.
- All AI deployment agreements contain express ownership and indemnity provisions covering outputs and training-data infringement claims.
- Your production workflows document the human creative contributions to AI-assisted outputs.
- Trade mark and design registrations are in place for commercially significant AI-generated visual assets.
- Your legal team has reviewed the current state of Singapore High Court authority on originality and authorship.
- Cross-border exploitation strategies account for the divergent treatment of AI-generated works in each target jurisdiction.
Frequently asked questions
Q: Can an AI system hold copyright in Singapore?
A: No. Singapore's intellectual property legislation requires a human author to subsist for copyright protection. An AI system cannot be a legal person and therefore cannot own or exercise copyright. Ownership must vest in a human individual or a legal entity that satisfies authorship criteria under Singapore law.
Q: Who owns copyright in an AI-generated work in Singapore?
A: This question remains unsettled under current Singapore intellectual property legislation. Potential claimants include the developer of the AI system, the user who supplied creative prompts, or the commissioning party under a contract. In practice, ownership is most reliably established through express contractual assignment before a dispute arises, rather than relying on statutory default rules.
Q: Does using copyrighted data to train an AI model infringe copyright in Singapore?
A: The position is genuinely contested. Singapore's intellectual property legislation does not contain a general text-and-data-mining exception comparable to those in some other jurisdictions. Reproducing protected works during training may constitute infringement unless a fair-dealing or other permitted-use defence applies. Businesses should conduct a rights-clearance review before commencing AI training at scale.
About Ferraz & Whitmore
Ferraz & Whitmore is an international law firm based in Lisbon, advising business clients across 46 jurisdictions. Our AI and technology law practice supports technology companies, financial institutions. Additionally, regional holding structures in Singapore and across the Asia-Pacific and Middle East region on AI-generated works. Intellectual property ownership, technology licensing, digital services regulation, and algorithmic accountability. We combine Portuguese civil law expertise with English common law tradition. giving clients a dual perspective that is directly relevant to Singapore's common law-based IP system and its interaction with civil law jurisdictions across Asia and the Gulf. Our attorneys have advised on technology licensing and IP structuring matters before the Singapore High Court and in SIAC arbitration proceedings. Engaging a lawyer in Singapore with cross-border experience in both common law and civil law systems is particularly valuable when AI-generated assets are deployed across multiple jurisdictions simultaneously. As an international law firm advising on Singapore matters, Ferraz & Whitmore brings together IP strategy, regulatory compliance, and cross-border transaction experience to support clients at every stage of their AI deployment. To discuss your specific situation, 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.