AI and Intellectual Property: Who Owns What AI Creates in 2026

You asked an AI to write your marketing copy. Another tool generated your logo. Your developer used Copilot to write half the codebase. Now a competitor is selling something almost identical and you want to sue. Here’s the problem: you might not own any of it.

Intellectual property law was not built for a world where machines create. It assumes a human made something. The moment that assumption breaks down — which AI does, constantly — the legal ground gets very shaky very fast. I’ve been tracking how UK law is responding to this, and the answer is: slowly, and with a lot of uncertainty baked in.

This article covers who actually owns AI-generated content under UK law, what the courts have decided so far, what happens with training data, and what steps UK businesses should take right now to protect themselves.

Who Actually Owns AI-Generated Work in 2026?

The short answer is: it depends on how much a human was involved. Copyright law in the UK, governed by the Copyright, Designs and Patents Act 1988, requires a human author. Always has. An AI system cannot be the legal author of anything — it has no legal personality and cannot hold rights.

But the UK has a quirk that most countries don’t. Section 9(3) of the CDPA says that for “computer-generated works” where there is no human author, the person who made the “necessary arrangements” for the work to be created gets the copyright. That’s the person who set up the system, wrote the prompts, chose the inputs.

In theory, this protects AI-generated content in the UK in a way it isn’t protected in the US or EU. In practice, the courts haven’t fully tested it yet. Nobody knows exactly what “necessary arrangements” means when you type “make me a logo in a blue gradient” into Midjourney.

When I looked into how the Intellectual Property Office is advising businesses on this, their own guidance admits the law “was not designed with AI in mind.” That’s as honest as government documents get.

The UK’s Official Position (And Where the Gaps Are)

The UK IPO ran a consultation on AI and IP that concluded in 2022. The government’s response confirmed it would not remove the computer-generated works provision from the CDPA — so Section 9(3) stays. That’s the good news.

The bad news is everything else remains murky. The IPO acknowledged that current law may not be fit for purpose as AI becomes more capable, and said it would “continue to monitor developments.” That is not reassuring if you are building a business on AI-generated assets right now.

What this means in practice: if you use AI tools and can demonstrate you made the meaningful creative choices — the direction, the selection, the refinement — you have a reasonable claim to copyright under Section 9(3). If you typed three words into a generator and clicked download, your claim is much weaker. The more creative input you contribute, the stronger your position.

What the Courts Have Actually Decided

The US case everyone cites is Thaler v Vidal, where inventor Stephen Thaler tried to list an AI system called DABUS as the inventor on a patent. The Federal Circuit said no — US patent law requires a human inventor. Full stop.

Thaler also tried to get copyright protection for an AI-generated image in the US. The US Copyright Office denied it repeatedly, and courts upheld that decision in 2023. The ruling was direct: copyright protects human expression. An image produced autonomously by a machine is not human expression.

The UK hasn’t had an equivalent landmark case yet. That’s not good news — it means there’s no definitive test of how Section 9(3) applies to modern generative AI. Until a case reaches the courts, businesses are operating on legal best guesses.

What has happened in the EU is worth watching. The EU AI Act, which came into force in 2024, includes transparency requirements — AI-generated content must be labelled as such in many contexts. That’s not copyright, but it signals a direction. The EU is treating AI content as a category that needs its own rules.

Training Data: The Copyright Bomb Nobody Talks About

Here’s the issue that actually keeps IP lawyers up at night. Generative AI models are trained on vast datasets scraped from the internet — images, text, code, music, everything. A lot of that content is copyrighted. Nobody asked the original creators.

Multiple lawsuits are running simultaneously:

• Getty Images sued Stability AI in the UK High Court in 2023, alleging it scraped millions of Getty photos to train Stable Diffusion without a licence. That case is still grinding through the system.
• A group of authors including John Grisham and George R.R. Martin sued OpenAI in the US, claiming their books were used to train GPT models without consent.
• The New York Times sued OpenAI and Microsoft, alleging wholesale reproduction of its journalism.
• In code: GitHub Copilot was sued by a group of developers over training on public repositories without attribution.

None of these have fully resolved. But the outcomes matter enormously. If courts decide training on copyrighted data is infringement, then every AI-generated output might be considered a derivative work — with potential liability flowing back to the tool providers and, in some interpretations, to the businesses using them.

I’ve seen some businesses quietly start asking their AI vendors for indemnification clauses — language that says the vendor will cover legal costs if a training data claim comes up. That’s a smart move.

What AI Companies Say You Own (And What They Don’t)

Terms of service vary wildly. Worth reading carefully.

Midjourney’s terms say you own the images you generate if you’re a paid subscriber. Free users give Midjourney a licence to use their outputs. But Midjourney also notes that it cannot guarantee third-party rights are clear — meaning a generated image might look similar to something a human artist made, and Midjourney won’t defend you if that artist sues.