Explain why nobody can give a single answer to who owns AI-generated content.
Open one piece you've made with an AI tool in the past month and write two things next to it: the choices you made yourself, and the choices you handed to the tool. Store the note somewhere you'll find it later. Lesson 4.3, "Working rules for AI content at work", sets out a method for keeping these records.
You need this note because no one can yet tell you for certain who owns work made with AI help. If you ask "who owns this AI image?", the answer can depend on the country, the kind of work and the year you ask. Early rulings and guidance keep looking at how much of the work came from a person. You are the only one who knows that about your own work.
Kelvin (an invented example, as is everything that happens to him) handles marketing for his family's bakery in Joo Chiat, which sells only in Singapore. Last month he used an image generator to make a poster for the Mid-Autumn mooncake launch. It showed lanterns, a moon over shophouse rooftops and a box of mooncakes in the foreground, and it looked great. He picked the subject and the mood. The generator settled the composition, the brushstrokes and the exact colours of the lanterns.
A little later, a friend who runs a print shop asked whether she could use a similar style for posters of her own. Kelvin had no idea whether the image was his to object over, or whether anyone owned it at all. He put the question to an AI assistant and got a two-line reply with no doubt in it. The question is still open, so a reply that certain was the part he should not have trusted.
He was honest with his friend and said he didn't know, and doubted that anyone did yet. Since then he has watched more carefully how much of each piece he makes himself.
Copyright means the rights the law gives a human author over copies of their work, for a period of time. Most countries built their copyright law around one simple case: someone writes a book, takes a photo or composes a song. The law assumes that a human made the creative choices behind the work.
With an AI tool, those choices are spread across several parties. You make some as the user. The generator makes others. The company that built the generator and the artists whose work the model was trained on also have a part. In Kelvin's case, that last group runs to thousands of artists.
In several countries, courts and regulators are working out how current law applies to AI, and a few have offered early views. The US Copyright Office registers only material created by a human, and anyone applying must disclose any AI-generated content. Other countries are handling it differently, some through court cases and some through consultations on new rules. For the position in Singapore, check the Intellectual Property Office of Singapore (IPOS) for current guidance.
Each country has its own copyright law, and international treaties align those laws only in broad terms. Singapore, the United States, the United Kingdom, China and the European Union might each take a different view of the same image. Kelvin's poster is on Instagram, where people in any country can see it. If he printed the image on a tote bag and sold it from an online shop that ships abroad, another country's law could apply alongside Singapore's.
For a social media post, or a slide deck that stays inside your company, the question of which country's law applies seldom causes trouble. It starts to matter when you sell something, license work to a client, or want to stop someone else from copying it.
Early decisions and guidance keep coming back to one factor: how much human creative input went into the finished result. Under a law that requires a human author, the parts you made yourself stand on firmer ground than the parts the tool generated. Kelvin could have produced his poster in three ways (also invented):
He types "Mid-Autumn poster for a bakery", takes the first image and stops there. Hardly any of it is his own work. He generates thirty images and picks one. Then he repaints the mooncake box by hand, adds his own photo of the bakery's shopfront and arranges the text himself. His work sits on top of AI output, in the middle ground where the law is least settled. He draws the whole poster himself and uses AI only to suggest colour palettes. The poster is clearly his.
You don't have to know exactly where the legal line falls. Your own input counts, and that is enough reason to write it down, as in the note you started with.
AI assistants often give neat, simple answers to questions that are not settled. They are trained to sound fluent and helpful, and a hedged answer sounds less helpful than a firm one. Lesson 6.1, "Why a fluent answer can still be false", in the course "AI fundamentals: what it is, how it works, where it fails", explains why. Be just as careful with confident blog posts, social media threads and colleagues who mean well.
Treat any one-line answer to an AI copyright question as unreliable unless it comes from a lawyer who has looked at your specific situation and the law of the relevant country. This course gives you working knowledge, enough to spot risks and ask good questions. It is not legal advice. If you plan to sell or license something, a lawyer's view is worth paying for.
Next, look back over the AI-assisted things you've made recently in the same way: the posters, the reports, the captions, the code. Next to each one, write what you decided yourself and what you left to the tool.
List three pieces of work you made with AI help. Next to each, note how much of the creative input was yours.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).