TRW & Co · Intellectual property

Can AI Be an Inventor or Author in Singapore? IPOS Consultation Explained

Updated 20 October 2026Legal insightSingapore10 min read
Typewriter printing the words Artificial Intelligence, evoking questions of AI and IP law in Singapore
Photo: Unsplash.

Under AI and IP law in Singapore, only a natural person can be an inventor and only a human can be an author. AI-assisted work is protected only where human contribution is sufficient. A MinLaw and IPOS consultation running from 26 August to 22 October 2026 asks how that test should apply, and how the AI training exception and output liability should work.

Key pointThe consultation is not about letting AI own rights; it is about where the line of sufficient human contribution sits. Businesses that document who framed problems, chose among outputs and edited results will be protected whichever way the law develops.

At a glance

01

Human inventors only

The Patents Act 1994 treats the inventor as the actual deviser, read as a natural person; IPOS rejected the DABUS application naming an AI system as inventor.

02

Human authors only

Under the Copyright Act 2021 an author must be a human; AI-assisted works are protected only where sufficient human intellectual effort is reflected in the work.

03

Training exception

The computational data analysis exception permits copying for AI training, including commercial use, provided the works were lawfully accessed; contract terms excluding it are void.

04

2026 consultation

MinLaw and IPOS launched a public consultation on 26 August 2026 covering copyright and patents; submissions close at 5pm on 22 October 2026.

05

Key open questions

Sufficient human contribution, the lawful access condition, liability for infringing AI outputs and how AI-generated prior art affects patent examination.

Swipe or scroll →

The short answer: can AI be an inventor or an author?

Under current AI and IP law in Singapore, no. Only a natural person can be named as the inventor on a Singapore patent, and only a human being can be the author of a work protected by copyright. Work produced with AI can still be protected, but only where a human made a sufficient creative or inventive contribution that is reflected in the result.

That position is now under review. On 26 August 2026 the Ministry of Law (MinLaw) and the Intellectual Property Office of Singapore (IPOS) launched a public consultation on artificial intelligence and Singapore’s intellectual property regime. It closes at 5pm on 22 October 2026. The consultation does not propose that AI systems should become inventors or authors. Instead it asks how much human contribution should be enough, how the computational data analysis exception for AI training should work, who is liable when AI output infringes, and how a flood of AI-generated disclosures could affect patent examination.

This article explains the current rules, the questions on the table and what businesses that build, train or use AI should do now.

Boardroom overlooking Marina Bay where legal teams review AI and IP law in Singapore
The 2026 consultation asks how much human input should be enough to protect AI-assisted inventions and works.

Patents: why only a human can be an inventor

The Patents Act 1994 defines an inventor as the actual deviser of the invention. The consultation paper explains that Singapore courts read this as the natural person who came up with the inventive concept, citing cases including the Court of Appeal's decision in Cicada Cube Pte Ltd v National University Hospital (Singapore) Pte Ltd [2018] SGCA 52. Applicants must file a statement identifying the inventors; if they do not, the application is treated as abandoned.

IPOS applied that rule to the well-known DABUS applications, in which an AI system was named as the inventor. The consultation paper records that IPOS rejected the application because only natural persons can be inventors under the Patents Act, which is consistent with the approach in most jurisdictions.

Two practical consequences follow. First, inventorship is a personal status that cannot be assigned, even though ownership of the patent can be. Second, a patent granted to someone not entitled to it can be revoked. Getting inventorship wrong on an AI-assisted invention is therefore not a paperwork error; it can undermine the patent itself.

What the consultation asks about inventorship

The consultation does not question the natural-person rule. It asks how human contribution should be assessed when AI is used in the inventive process: for example, where a researcher formulates the problem, selects among AI-generated candidate solutions or modifies an AI output. It tests these questions against four hypothetical scenarios and asks what information about the AI model applicants should disclose.

The paper sets out four possible ways forward on clarification: maintaining the current position and letting case law develop, legal refinements, practical information resources, or non-binding technical measures. It also asks a separate set of questions about AI-generated prior art, including whether large volumes of machine-generated technical disclosures should count when assessing novelty and inventive step, and how examiners could search them.

Copyright: the human authorship requirement

Under the Copyright Act 2021, literary, dramatic, musical and artistic works are protected as authorial works. The consultation paper confirms that, under Singapore law, an author must be a human being. A work involving AI is protected only if a human made it and it is an authorial creation causally connected with the engagement of the human intellect. The paper cites the Court of Appeal's observation in Asia Pacific Publishing Pte Ltd v Pioneers & Leaders (Publishers) Pte Ltd [2011] 4 SLR 381 that copyright aims to encourage the creativity of natural authors.

Courts look at whether the human's acts and decisions show sufficient intellectual effort and are reflected in the final work. Mechanical tasks and fact-checking do not count. That leaves a wide grey area for generative AI: a one-line prompt is unlikely to be enough, while substantial human selection, arrangement and editing of AI outputs may well be.

The consultation questions on authorship

The consultation asks which creative acts should count. It distinguishes pre-generation acts, such as choosing the model, writing detailed prompts, iterating and setting technical parameters, from post-generation acts, such as selecting outputs, stopping generation, arranging or compiling outputs and editing them. It treats editing as the most direct form of human intervention. It also asks how agentic AI, which can take multi-step actions with little human direction, changes the analysis, and whether sufficiency principles should come from the courts, legislation or non-binding guidance.

Training AI: the computational data analysis exception

Singapore already has one of the more permissive copyright exceptions for AI training. Sections 243 and 244 of the Copyright Act 2021 permit copying for computational data analysis, which includes using a work as an example of a type of information or data to improve a computer program. The consultation paper notes that the exception covers generative AI and commercial uses: none of its conditions turns on whether the use is commercial.

There are safeguards for rights owners. The most important is lawful access: the user must have accessed the works lawfully. The statutory illustrations treat circumventing access controls, and accessing a database in breach of its terms of use, as examples of unlawful access. The Act also makes void any contract term that excludes or restricts the exception.

That combination creates tension. If breaching website terms makes access unlawful, but terms that restrict the exception are void, which prevails? The paper flags the uncertainty and asks whether clarity should come from case law, legislation or non-binding instruments, including a possible deeming provision for contract breaches. It also confirms that tools such as robots.txt do not, in themselves, create legal obligations or count as access controls.

Practical point

The computational data analysis exception does not remove the need for diligence. AI developers should be able to show where training data came from and how it was accessed, and keep that record with each model version. Rights owners who want to limit training on their content should focus on genuine access controls and well-drafted access terms, because a contractual ban on training alone may be void under AI and IP law in Singapore as it stands.

Who is liable when AI output infringes?

The consultation also asks how existing copyright principles apply when AI-generated output reproduces protected material. Possible infringers include the developer who trained the model, the deployer who offers it to customers and the end user who prompts it. The paper refers to the established law on reproduction, communication to the public and authorising infringement, and asks how agentic AI affects that analysis.

It further asks whether technical measures to reduce infringing output should be expected, and whether expectations should scale with the size of the developer, deployer or user. Businesses deploying third-party AI tools should watch this closely, because the answer will shape what they can reasonably demand from vendors in warranties and indemnities.

Current position and consultation questions at a glance
IssueCurrent positionWhat the consultation asks
AI as inventorOnly a natural person can be an inventor; IPOS rejected the DABUS applicationHow to assess human contribution to AI-assisted inventions and what to disclose about the model
AI-generated prior artPrior art rules apply generallyWhether and how large-scale AI-generated disclosures should affect novelty, inventive step and searching
AI as authorAn author must be a human beingWhich human acts before and after generation should count as sufficient creative contribution
Training on copyright worksComputational data analysis exception applies, subject to lawful access; contract overrides voidWhether scope and the lawful access condition need clarification, and how to support technical measures
Infringing outputsGeneral principles on reproduction, communication and authorisation applyHow liability is shared among developers, deployers and users, and expected safeguards

What AI and IP law in Singapore means for different businesses

The same rules land differently depending on where a business sits in the AI value chain. The current position, and the questions in the consultation, raise distinct issues for each group.

  • AI developers and model trainers. The computational data analysis exception is the key asset. Its value depends on being able to show lawful access to training data, so provenance records and access terms matter more than broad assertions that training is permitted.
  • Deployers and platforms. Businesses that offer AI tools to customers face the output liability questions most directly. Vendor contracts, usage policies and output filters are the practical levers while the law develops.
  • R&D-intensive companies. Pharmaceutical, materials, engineering and software businesses that use AI in research need disciplined inventorship records so that AI-assisted inventions remain patentable under AI and IP law in Singapore.
  • Content and media businesses. Publishers, agencies and creators need to know when AI-assisted content is protected, and how to stop others training on their catalogues where they wish to license it instead.
  • Investors and acquirers. Due diligence on AI businesses should test data provenance, inventorship records and the ownership of AI-assisted outputs, because gaps in any of these can reduce the value of the IP being acquired.

For each group, the most useful response to the uncertainty in AI and IP law in Singapore is the same: build records now that would satisfy a demanding test of human contribution and lawful access later.

What businesses should do now

Whatever the consultation produces, any change to AI and IP law in Singapore will take time to arrive. Meanwhile, the current rules already reward businesses that document human contribution and data provenance. We recommend the following steps.

  1. Map where AI touches your IP. List the products, research projects and content workflows that use AI, and identify which outputs you expect to protect by patent or copyright.
  2. Record human contribution. For inventions, keep lab notebooks or design records showing who framed the problem, chose among outputs and refined the solution. For content, keep drafts, prompts and editing histories that show human creative choices.
  3. Settle inventorship and ownership early. Name natural-person inventors accurately and make sure employment, consultancy and collaboration agreements assign rights in AI-assisted work to the right entity.
  4. Document training data provenance. Record the sources of training data, how access was obtained and the terms that applied, so you can rely on the computational data analysis exception with confidence.
  5. Review vendor contracts. Check AI vendor terms for ownership of outputs, warranties on training data, infringement indemnities and limits on vendor use of your inputs.
  6. Consider responding to the consultation. If your business has a strong view, a short, evidence-based submission before the 22 October 2026 deadline can help shape the outcome.

An illustrative scenario

This scenario is illustrative only and does not describe any real client.

A Singapore materials company uses a machine learning model to propose thousands of candidate polymer formulations. A research scientist defines the target properties, filters the candidates using her own experimental knowledge, tests a shortlist and then modifies the most promising formulation to overcome a stability problem the model did not anticipate.

Under current AI and IP law in Singapore, the scientist is likely to be the right person to name as inventor, because the inventive concept that solves the stability problem came from her. The company strengthens its position by keeping clear records of each human decision. It also checks that the model was trained on data it was entitled to use, and that its research agreement with a university partner assigns rights in AI-assisted results. Had the scientist simply picked the model's top-ranked output without further contribution, the inventorship question would be much harder, which is exactly the kind of case the consultation is exploring.

Risks and common mistakes

Naming the wrong inventor

Some teams list the person who ran the AI tool, or everyone on the project, as inventors. Inventorship follows the inventive concept, not job titles or tool usage. Wrong inventorship can invite challenges to entitlement and, in serious cases, revocation of the patent.

Run an inventorship review before filing, and repeat it if claims change materially during prosecution.

Assuming AI output is automatically owned

Businesses often assume they own whatever their AI tools produce. If there is no sufficient human authorship, there may be no copyright for anyone to own, and competitors may be free to copy. Vendor terms may also allocate rights differently.

Where exclusivity matters, ensure meaningful human creative input and review the contractual position. Our article on copyright ownership decisions explains the wider ownership rules.

Relying on terms of use to stop AI training

Rights owners sometimes rely solely on website terms prohibiting scraping or AI training. Under the Copyright Act, terms that exclude or restrict the computational data analysis exception are void, although breach of access terms may still bear on whether access was lawful. The interaction is one of the uncertainties the consultation seeks to resolve.

Combine access controls, clear access terms and commercial licensing rather than relying on a single clause.

Ignoring cross-border differences

Other jurisdictions take different approaches to text and data mining exceptions, computer-generated works and AI-assisted inventions. A model trained lawfully in Singapore may face different rules when deployed elsewhere, and patent offices abroad may assess AI-assisted inventorship differently.

Plan filings and deployments market by market.

The cross-border picture

Singapore's consultation sits within a global debate. The paper notes, for example, that the United Kingdom has proposed abolishing protection for computer-generated works with no human author. Courts and patent offices in several jurisdictions have also rejected AI inventorship. Businesses operating across Asia, the Middle East and Europe need a strategy that works under the strictest relevant regime.

TRW & Co advises on AI and IP law in Singapore alongside our offices in London, Dubai, Dhaka and China. Related reading: our pages on copyright and content, patent registration and AI governance; our guide to registering a patent in Singapore; our analysis of the PDPC advisory guidelines on generative AI; and our article on software commercialisation risks. To discuss your position or a consultation response, contact TRW & Co. Where Singapore court or tribunal representation is required, we work alongside Singapore-qualified counsel.

Frequently asked questions

01Can an AI system be named as an inventor in Singapore?

No. The Patents Act 1994 defines the inventor as the actual deviser of the invention, which Singapore courts read as the natural person who devised the inventive concept. IPOS rejected the DABUS application, which named an AI system as inventor, on that basis. An AI-assisted invention can still be patented, but a human who contributed the inventive concept must be named. The 2026 consultation asks how that human contribution should be assessed, not whether AI should become an inventor.

02Is AI-generated content protected by copyright in Singapore?

Only if there is sufficient human authorship. The consultation paper confirms that under Singapore law an author must be a human being, and that a work involving AI is protected only where it is an authorial creation causally connected with the engagement of the human intellect. Courts look at whether human acts and decisions show sufficient intellectual effort reflected in the final work. Simple prompts are unlikely to be enough, whereas substantial selection, arrangement and editing may be.

03Is it legal to train AI on copyright works in Singapore?

Often yes. Sections 243 and 244 of the Copyright Act 2021 provide a computational data analysis exception that covers using works to improve computer programs, including generative AI and commercial uses. The main condition is lawful access to the works, and circumventing access controls or breaching database access terms may make access unlawful. Contract terms that exclude or restrict the exception are void. The consultation asks whether these rules need clarifying.

04When does the AI and IP consultation close?

The public consultation by the Ministry of Law and IPOS opened on 26 August 2026 and closes at 5pm on 22 October 2026. Submissions can be made through separate online forms for copyright and patents or by email to MinLaw. MinLaw may publish submissions, so any confidential material must be clearly marked and placed in a separate annex. Requests to keep all or most of a submission confidential will not be accepted.

05Will Singapore change its law on AI inventorship or authorship?

It is too early to say. The consultation asks open questions and sets out possible approaches, ranging from leaving matters to case law, to legal refinements, practical guidance and non-binding technical measures. It does not propose recognising AI systems as inventors or authors. Any legislative change would follow MinLaw and IPOS's review of responses, so businesses should plan under the current rules while monitoring developments.

06Who is liable if AI output infringes copyright?

Under current law, general copyright principles apply, so liability may fall on whoever reproduces or communicates the infringing material or authorises others to do so. Depending on the facts, that could be the AI developer, the deployer offering the tool or the end user. The consultation asks how responsibility should be shared and whether technical safeguards should be expected. Businesses should address the risk now through vendor warranties, indemnities and output review processes.

Sources and official materials

  1. MinLaw: Public Consultation on Artificial Intelligence and Singapore's Intellectual Property Regime — Consultation scope, dates and how to respond
  2. Public Consultation on AI and IP (consultation paper, 26 August 2026) — Current law, issues and options on copyright and patents
  3. Copyright Act 2021 — Authorship and the computational data analysis exception
  4. Patents Act 1994 — Inventorship, entitlement and revocation
  5. Intellectual Property Office of Singapore — IPOS guidance and patent and copyright resources

General information only, current as at 20 October 2026. It is not legal advice and does not create a lawyer–client relationship. Singapore law and regulatory practice change; check current official sources and take advice on the facts of your matter.

TRW on LinkedIn & Facebook

From the firm,
this week.

Conference notes, transactions in the news and the people we work with — follow TRW & Co for the full feed.

Call the Singapore desk+880 1779 127165 WhatsApp · message usMessage our team