You Paid for the AI. That Doesn’t Mean You Own
What It Made.
There is an assumption I keep encountering with businesses using generative AI. It usually goes something like this: we paid for the platform, we wrote the prompt, we generated the output, so we own it.
It feels intuitively right. You commissioned the machine, in a sense. You told it what you wanted. You may even be paying a fairly substantial subscription fee for the privilege.
But copyright doesn’t really care how much you paid for the software.
I recently spoke to theSun about this while discussing AI-generated content and the changes Malaysia may eventually need to make to its copyright framework. The more I thought about it, the more I found myself coming back to three questions.
Does copyright exist in the first place? If it does, who owns it? And even if you own something, is the finished output actually safe to use?
We tend to collapse those questions into one. AI makes that increasingly dangerous.
Before asking who owns it, ask whether there is anything to own
This is probably the strangest part of the problem for businesses.
You can have an image sitting on your screen. It might look completely original. It might be commercially useful. Your company might already be using it in a campaign.
And the first copyright question may still be whether copyright exists in all of it at all.
Copyright law was built around human creativity. Generative AI unsettles that fairly basic assumption because the distance between the human and the final expression can vary enormously.
At one end, a designer might develop an idea, use AI to explore possibilities, select elements from different outputs, redraw them, change the composition and spend hours shaping the final work.
At the other end, someone types “make me a logo for a fintech company with a lion” and chooses version three.
Calling both of those things “AI-generated content” isn’t particularly helpful. The creative processes are completely different.
This is why I think the distinction between AI-assisted and predominantly AI-generated work is going to matter.
The US Copyright Office has already been grappling with essentially this problem. Its position is not that using AI somehow poisons a work and prevents copyright. Human-created expression can still be protected.
The difficulty arises when the expressive elements themselves are generated by the machine without sufficient human creative control.
That seems sensible to me.
The interesting question isn’t whether AI was used.
It’s what the human actually did.
Three weeks of prompting may not help you
This is where things become uncomfortable for the idea of the “prompt engineer” as author.
Prompting plainly involves skill. A good prompt can produce a much better result than a bad one. Anyone who has spent time with these systems knows that.
But copyright isn’t a prize for effort.
Imagine two people.
The first spends three weeks prompting an image generator. Hundreds of prompts, endless variations.
Eventually the machine produces exactly the image they were hoping for.
The second spends an afternoon with the same system. But she takes several outputs, chooses particular elements, changes the composition, redraws parts of the image, adds material of her own and makes the decisions that determine what the final work actually becomes.
Who exercised more creative control over the final expression?
I’d be much more interested in the second person.
That doesn’t mean prompts are irrelevant. A sufficiently expressive prompt might itself contain copyrightable material. But owning the words you typed into a box and owning what the machine
subsequently produced are not necessarily the same thing.
The number of prompts isn’t the point.
The human hand is.
Then there is the small matter of what you actually bought
Suppose we get over that hurdle. There is enough human creativity in the finished asset for copyright to exist.
The business still doesn’t necessarily own it.
Perhaps an employee created the human elements. Perhaps it was a freelancer. Perhaps you hired an agency that used several AI systems internally and then had a designer finish the work.
These aren’t particularly futuristic problems. They’re old-fashioned IP ownership problems that happen to involve a new tool.
Contracts still matter.
What becomes more interesting is the relationship with the AI platform itself.
Businesses understandably read “commercial use” and hear “ownership”. But those concepts are doing different jobs.
The platform may give you contractual permission to use an output. That doesn’t answer whether copyright subsists in the output. It doesn’t necessarily tell you who owns the human-created portions. And it certainly doesn’t guarantee that the result won’t interfere with somebody else’s rights.
Access is not ownership. Permission is not ownership. And ownership isn’t necessarily freedom to use.
That last distinction is particularly important.
The machine gave it to you. Someone else may still have a problem
with it.
Imagine you’ve generated an image for an internal presentation. Perhaps the risk is tolerable.
Now imagine you’ve built your company’s visual identity around it. It’s on your packaging. You’ve registered domains around it. You’ve spent RM500,000 on a campaign. You’re about to take it into three new markets.
I would want to know considerably more about that image.
Does it reproduce recognisable expression from another work? Is the logo too close to an existing trademark? Does the packaging resemble somebody else’s protected design? Did somebody deliberately ask the AI to imitate a particular artist or commercial asset?
There is a huge legal fight happening internationally over the material used to train generative AI systems.
That matters enormously.
But the company deploying the output has a much more immediate concern. You’re the one putting it into the market.
AI hasn’t repealed due diligence.
In fact, I suspect it makes due diligence more important. We can now create brands, images, copy, music, software and product concepts at a speed that the traditional review process was never designed to handle.
Creation has accelerated. Legal and commercial clearance hasn’t.
That gap is where I expect a lot of future disputes to live.
Keep a record of the human hand
If human contribution is going to matter, businesses will eventually need to prove it.
I don’t mean keeping 700 screenshots of prompts in a folder and calling that an AI governance policy.
I’d rather know who developed the concept. What the AI produced. What was discarded. What the human changed. Who rewrote or redrew things. Who combined the elements. Who made the decisions that turned a machine output into the finished commercial asset.
This sounds bureaucratic until the asset becomes valuable.
Then imagine trying to reconstruct that history four years later because somebody has challenged your ownership.
You probably won’t remember which version came first. The employee who created it may have left. The agency may no longer have the working files. The AI model itself may have changed several times.
If the human contribution isn’t documented when the work is being created, rights that might have been perfectly defensible can become very difficult to prove later.
Keep a record of the human hand.
I think that will become a fairly important principle for businesses using generative AI.
Maybe copyright is the wrong place to stop
There is another habit from traditional IP practice that I think AI should force us to reconsider.
We like putting things into boxes.
Is this copyright? Is this a trademark? Is this a design?
Businesses don’t experience their assets that way. They just know that something has become valuable and they don’t want a competitor taking it.
If an AI-assisted asset sits in an uncertain area of copyright law, that doesn’t automatically mean it is unprotectable.
A brand can be protected through trademarks. A product’s appearance may qualify for industrial design protection. Confidential information can be controlled through contracts and, in the right circumstances, trade secret protection. Human-created components of a larger AI-assisted work may have copyright even if other components don’t. Contracts can determine who owns what between employees, freelancers and agencies.
This is why I think the more useful question for a business is not simply, “Do we own the copyright?“
It is:
“How do we control this asset?“
That changes the conversation.
Copyright becomes one layer of an IP strategy rather than the entire strategy.
Create. Clear. Control.
When I spoke to theSun about all of this, I eventually found myself reducing the practical advice to three words.
Create. Clear. Control.
Create with meaningful human creative involvement.
Clear AI-assisted outputs for third-party intellectual property risks before commercial use.
Control valuable assets through the appropriate combination of copyright, trademarks, designs, contracts and development records.
That’s the short version.
In practice, creating with meaningful human involvement means that if the asset matters to the business, you shouldn’t simply accept what the machine gives you. Shape it. Select. Reject. Edit. Rewrite. Combine.
Redraw. Make decisions about what the finished thing actually becomes.
Clearance should become more serious as the commercial stakes increase. If you’re about to build a brand around an AI-assisted asset, invest heavily in promoting it or take it into multiple markets, relying on the fact that an AI platform allowed you to generate it isn’t enough. Check the relevant IP landscape and understand the risks.
Then work out how you’re going to control it. Get the contracts right. Register trademarks or designs where they make sense. Keep sensible development records. Use copyright where copyright genuinely exists.
Build layers around the things that matter.
None of this solves the larger policy problem.
Malaysia still has decisions to make about how its copyright framework should deal with AI-assisted and predominantly AI-generated works, and about the much more contentious question of copyrighted material being used to train AI systems.
I think we should be ambitious about AI. There is enormous economic value to be created here.
But I’ve used the same phrase before and I still think it captures the balance fairly well.
Malaysia should be pro-AI without becoming pro-piracy.
Human creators need to remain part of the economic value chain. AI companies need rules they can actually operate under. And businesses need enough certainty to know what they own and what they can safely take to market.
Those questions won’t all be resolved tomorrow.
The commercial use of AI isn’t going to wait for them.
So for now, if your company is building something valuable with generative AI, I’d resist the temptation to assume that paying for the tool settled the ownership question.
It didn’t.
Ask what the human created. Work out who owns that contribution. Find out whether the finished asset is safe to use.
Then create, clear and control.
Author
Visharad Venugopal Mannadiar
Founder of Brandguard
Certified Intellectual Property Valuer (AMAVI)
About the author
Visharad is a certified IP valuer and intellectual property advisor focused on the intersection of artificial intelligence, intellectual property, and strategic defensibility.