By the WahLiao desk · Last verified 1 October 2026
In Singapore, copyright only protects work that comes from a human. If Super Intelligence, the kind of AI behind ChatGPT, Claude and Gemini, produces an image or a block of text with little more than a prompt from you, that output may not be protected by copyright at all, so nobody can stop others copying it. What the tools do promise is contractual: OpenAI, Anthropic, Google and Microsoft all say they do not claim ownership of what you generate, and OpenAI and Anthropic assign to you whatever rights they may have. No Singapore court has yet ruled directly on AI output, and the Ministry of Law (MinLaw) and IPOS opened a public consultation on AI and intellectual property on 26 August 2026, closing on 22 October 2026. The rules may be clarified, but they have not changed yet.
The single most useful rule: the more of the creative work that is yours, the safer you are. Write, edit, select, arrange and redraw, and keep a record of what you did. A raw output you pasted in untouched is the weakest thing you can own and the easiest to get into trouble with.
AI and copyright in Singapore: quick facts
| The law | Copyright Act 2021; no AI-specific rules on ownership of outputs yet |
| Who can be an author | A human. The Court of Appeal requires an authorial creation causally connected with the engagement of the human intellect |
| Pure machine output | May not be protected by copyright; not yet tested in a Singapore court |
| Training on others’ works | Section 244 allows copying for computational data analysis, but only with lawful access, and it does not cover what the model later produces |
| What the tools say | OpenAI, Anthropic, Google and Microsoft do not claim ownership of your outputs |
| Legal cover | Paid business plans from all four offer some protection against copyright claims; free and consumer plans do not |
| Under review | MinLaw and IPOS public consultation, 26 August to 22 October 2026 |
The WahLiao Verdict
| Prompt only | Treat it as unprotected. Anyone may be able to copy it. |
| Your edits | This is where your copyright lives. Keep drafts. |
| Logos and brands | Do a trade mark search before you print anything. |
| Real faces and voices | Do not use them without permission. Ever. |
| Students | Declare it where it is allowed. Do not use it where it is not. |
Why a human has to be the author
Singapore’s copyright law has always been built around people. In Asia Pacific Publishing v Pioneers & Leaders (2011), the Court of Appeal said the objective of copyright law is to encourage the creativity of natural authors. In Global Yellow Pages v Promedia Directories (2017), it went further: for copyright to exist, there must be an authorial creation that is causally connected with the engagement of the human intellect, meaning intellectual effort, creativity, or the exercise of skill and judgment directed at creating the work itself. Neither case was about Super Intelligence, but MinLaw’s August 2026 consultation paper applies the same test: a work made with AI involvement is protected only if it is made by a human and meets that standard.
What that means in practice is still open. The consultation paper sets out three kinds of human contribution it is asking about: choices before generation (which model, how detailed the prompts, how many rounds, which settings), selecting and arranging outputs, and editing afterwards (rewriting, redrawing, retouching). It notes that major jurisdictions have been more willing to recognise protection where a human has edited or transformed the output. Until the government or a court says more, assume a single prompt earns you little, and your own rewriting and redrawing earns you the most.
What the big tools promise you
Terms of use cannot create copyright where the law gives none. They can only settle things between you and the provider. With that caveat, here is what each says as of 1 October 2026.
| Provider | Who owns the output | Business indemnity |
|---|---|---|
| OpenAI (ChatGPT) | You own it, to the extent the law allows; OpenAI assigns its rights to you. Similar output may go to other users | Business and API customers are covered for third-party IP claims that the services infringe, with exclusions |
| Anthropic (Claude) | Anthropic assigns you its rights, if any, provided you follow the terms | Paid commercial customers are covered for IP claims over authorised use and outputs; excludes trade mark use in commerce and outputs you modify |
| Google (Gemini) | Google says it won’t claim ownership of original content you generate | Google Cloud and Workspace generative AI indemnity, since October 2023, if you did not try to infringe |
| Microsoft (Copilot) | Microsoft does not claim ownership of what you receive from its AI services | Customer Copyright Commitment for paid commercial Copilot services, if you use the built-in guardrails |
Two points matter for small businesses. First, the protection is on business plans, not the free or personal ones, so a sole proprietor using a personal ChatGPT Plus or Claude Pro account has no indemnity. Our free vs paid plans guide has the prices. Second, every indemnity has exclusions: output you changed, material you fed in yourself, and uses you knew or should have known would infringe. They are a safety net, not permission.
Training: the section 244 exception
Singapore was early in this area. Since 2021, section 244 of the Copyright Act has let people copy works for computational data analysis, which MinLaw confirms covers training AI models, whether commercial or not. Terms in a contract that try to exclude or restrict it are void under section 187. There is a firm condition: you need lawful access. MinLaw’s paper says you cannot rely on the exception if you breach a database’s terms of use or get around a paywall or other access control. For how training works, see How Super Intelligence works.
The exception stops at training. It does not cover the eventual use of the trained model, so if a tool produces something that copies a protected work, the ordinary rules on infringement apply. How responsibility should be split between the developer, the business deploying the tool, and you as the user is one of the questions the current consultation asks.
The real risks: copies, brands and faces
Output that looks like someone else’s work. If an image closely resembles a known character or a photographer’s picture, or text reproduces a passage from a book or article, using it can infringe regardless of how it was made. Asking for something “in the style of” a named living artist raises the odds. Run a reverse image search before you publish anything commercial.
Trade marks. Image tools will happily produce logos that resemble existing brands. A logo you cannot register or that clashes with someone else’s mark is a costly mistake once it is on your signboard and packaging. Anthropic’s commercial indemnity, for one, specifically excludes trade mark claims arising from using an output in trade.
Real people. Putting a real person’s face or voice into generated content, especially in adverts or anything that could mislead, brings a different set of laws into play. Our page on deepfakes and the law covers what is prohibited and where to report, and deepfake scams covers how to spot them.
What you upload. Pasting a client’s brief or another company’s material into a chatbot raises confidentiality and data questions too; see what happens to what you type.
Practical rules for businesses, students and creators
Small businesses. Use Super Intelligence for drafts, then rewrite in your own voice. For logos and brand assets, treat the output as a sketch and have a designer finish it, and check it against the IPOS trade mark register. If you depend on the tools for client work, a business plan with an indemnity is worth the extra cost. Keep prompts, drafts and edits, so you can show your human contribution later. Grants that can help with the cost of tools are in our SME grants guide.
Students. Copyright is rarely the issue; school rules are. In a 6 May 2026 parliamentary reply, MOE said that where secondary schools design tasks that allow AI, students must state where they have used AI and cite their sources, and that AI is prohibited in national exams. At NUS, instructors decide whether AI is allowed in an assessment, and using AI output without acknowledgement is treated like copying a source without citation. Check your own school’s or university’s policy for each assignment. More in Children and Super Intelligence.
Creators selling AI art. Be honest with buyers and platforms that the work is AI-assisted. Do not promise exclusive rights you may not have: a buyer paying for a “unique” design may find the underlying image unprotected and similar output available to others. Sell your finished, edited work, not raw generations, and read the terms of the marketplace you sell on.
How the US compares
The United States is further down the same road. The US Copyright Office concluded in January 2025 that prompts alone are generally not enough for copyright, because they work as instructions conveying unprotectable ideas. In Thaler v Perlmutter, the courts held that a work made entirely by an AI system cannot be registered, and the US Supreme Court declined to hear the appeal on 2 March 2026. Singapore’s human-authorship case law points in a similar direction, but US rulings do not bind Singapore courts.
AI copyright in Singapore: FAQ
Do I own the copyright in images I make with AI in Singapore?
Possibly not, if all you did was write a prompt. Singapore copyright requires human authorship, and no court here has yet decided how much human input is enough. Your own edits, selection and arrangement are the parts most likely to be protected.
Can I use ChatGPT or Claude output commercially?
Generally yes. OpenAI and Anthropic assign their rights in outputs to you, subject to their terms. That does not guarantee the output is free of someone else’s copyright or trade mark, so check before you publish or print.
Is it legal to train AI on copyrighted material in Singapore?
Section 244 of the Copyright Act 2021 allows copying for computational data analysis, including AI training, if you have lawful access to the material. It does not cover infringing outputs, and MinLaw and IPOS are consulting on how it should apply until 22 October 2026.
Can I copyright a logo designed by AI?
Copyright is uncertain for a purely generated logo. A trade mark registration with IPOS is the more practical protection, but first check that the design does not resemble an existing mark, and ideally have a designer rework it.
Do students have to declare AI use in Singapore schools?
Yes, where it is allowed. MOE says students must state where they used AI and cite sources, and AI is prohibited in national exams. Universities such as NUS require acknowledgement of AI use and leave permission to each instructor.
Read next
This page belongs to Super Intelligence. Next, read Using Super Intelligence well.
Sources checked 1 October 2026: MinLaw, Public Consultation on AI and Singapore’s IP Regime; MinLaw and IPOS, consultation paper (26 August 2026); Court of Appeal, Global Yellow Pages v Promedia Directories [2017] SGCA 28; OpenAI, Terms of Use; OpenAI, Services Agreement; Anthropic, Consumer Terms; Anthropic, Commercial Terms; Google, Terms of Service; Google Cloud, generative AI indemnification; Microsoft, Services Agreement; Microsoft, Customer Copyright Commitment; MOE, parliamentary reply on AI usage in schools (6 May 2026); NUS, Policy for Use of AI in Teaching and Learning; Mayer Brown, Supreme Court denies review in Thaler v Perlmutter. General information, not legal advice. Last updated 1 October 2026.
