There are plenty of ways to make a patent office have a bad afternoon. Apparently, writing the name of a machine in the box marked “inventor” is an especially effective one.
That’s what happened with DABUS, an artificial intelligence system developed by Stephen Thaler. Two proposed inventions, a food container and an attention-grabbing light, became the center of a legal argument that traveled through patent offices and courts around the world. The objects themselves sound almost aggressively ordinary. A container! A flashing light! We could be discussing the contents of a kitchen drawer.
Instead, we’re discussing who gets to invent the future. Which is a lot of responsibility for a food container.
Thaler’s position was that DABUS generated the inventions autonomously, and that he should be named as the applicant and owner rather than take credit as their inventor. Several major patent systems rejected that arrangement. Their decisions weren’t a scientific verdict that computers can never produce useful ideas. They addressed who qualifies as an inventor under existing law.
That distinction matters throughout this story. So does another: a court discussing an applicant’s account of machine creativity hasn’t necessarily verified that account through an independent experiment. DABUS hasn’t been handed a certificate proving it has imagination, feelings, or a little internal voice complaining about Mondays.
The real story is strange enough without adding those things. It involves credit, ownership, a remarkably stubborn legal campaign, and a question we’re going to keep encountering as software becomes more capable: when a tool contributes something unexpected, how do we describe what actually happened?
What exactly is DABUS?
DABUS is a system Thaler presents as capable of generating inventions. The name has been expanded differently in different records: the American appellate opinion uses “Device for the Autonomous Bootstrapping of Unified Science,” while other accounts use “Unified Sentience.” Either version sounds like something you should avoid accidentally switching on while looking for the bathroom light.
The important part is Thaler’s claim about its role. He described the machine as the source of the inventive concepts, rather than a program carrying out a fully formed design he had already supplied. That became the premise of the applications and the test cases.
We should treat that as a claim about this particular system, not a shortcut to declaring that all AI is independently inventive. Different programs do different things. A model suggesting sentences, software optimizing a bracket, and a research system searching possible chemical structures aren’t interchangeable simply because we put “AI” on all three boxes.
We also need to separate generating an output from establishing that the output is useful, original, and technically workable. A machine can produce something surprising that fails immediately when somebody tries to build it. So can a person. Anyone who has ever confidently assembled furniture backward knows that confidence and functionality are very separate departments.
For DABUS, the legal campaign deliberately pushed the difficult version of the question. It wasn’t merely “Can someone patent something developed with software?” It was “What happens if the applicant says there was no person who qualifies as the inventor?”
A container and a light walk into a legal battle
The applications associated with DABUS concerned a food container with a fractal surface and a device or method for attracting attention using light. In the United States, the appellate record refers to the proposed inventions as the “Fractal Container” and “Neural Flame.”
A fractal pattern involves structure repeated at different scales. Think of the way a branching shape can echo itself in smaller branches, although a manufactured object necessarily has practical limits. The container proposal involved a textured geometry rather than a plain, smooth exterior. The second proposal concerned a light signal designed to attract attention.
Neither description means these objects are products you can order, that their performance has been independently established here, or that every claim would survive a full patent examination. A patent application is a request for a particular legal protection. It isn’t a consumer review, a safety certification, or a standing ovation from science.
I actually like that the objects are so unglamorous. We tend to imagine machine invention arriving with a cure for everything or a spaceship that folds the universe like a fitted sheet. DABUS arrived in the legal conversation carrying a container. A small object can expose a very large problem if you put it in the right paperwork.
And the paperwork asked for a name.
The inventor and the owner are different jobs
This is where the story gets much easier to follow. The inventor is the person credited under the relevant law with making the inventive contribution. The owner is whoever holds the patent rights. Those can be different people or entities.
A company can own a patent without being the inventor. An employee might be named as an inventor while rights belong to an employer through an agreement or applicable law. Naming an inventor doesn’t automatically tell you who gets the money, controls a license, or decides whether to sue.
Thaler wasn’t asking courts to give DABUS a bank account so it could collect royalties and buy a vacation home. His campaign sought to list the machine as inventor while placing ownership with him. That still created a problem: what legal path moved the right to apply for a patent from a machine to its owner?
Owning the equipment doesn’t answer every question about its output. If I own a camera, that fact alone doesn’t settle every copyright question involving photographs taken with it. If I own a restaurant’s oven, that doesn’t magically make me the inventor of everything that comes out of it. Equipment ownership and creative contribution describe different relationships.
DABUS forces those relationships into view. It asks whether existing rules can accommodate an applicant who says the inventive contribution happened inside a system that cannot hold and transfer rights in the ordinary way.
Why the United States said no
In August 2022, the U.S. Court of Appeals for the Federal Circuit ruled in Thaler v. Vidal that an inventor under the Patent Act must be a natural person. The court treated the statutory wording, including its use of “individual,” as decisive.
The decision didn’t establish that an invention becomes unpatentable the moment AI appears anywhere in its development. It addressed applications naming DABUS as the sole inventor. That is a narrower question, with a much broader public reputation.
Imagine a researcher using software to compare designs, then conceiving a specific solution. That situation calls for an assessment of the researcher’s contribution. Now imagine an applicant saying, “The machine conceived this, and I didn’t.” Those accounts raise different inventorship issues even if both involve a computer.
The American ruling left the debate about changing the statute to the people who can change statutes. Courts interpret the legal framework before them; they don’t get unlimited freedom to redesign it because a new technology makes the old categories uncomfortable.
There’s a temptation to summarize this as law refusing to keep up. Sometimes legal rules do need revision. But first we should understand what the rule is doing, whose interests it protects, and what would happen if we replaced it. “The computer did something impressive” doesn’t supply all those answers.
The British decision asked an ownership question too
In December 2023, the UK Supreme Court unanimously dismissed Thaler’s appeal. Under the Patents Act 1977, DABUS couldn’t be named as the inventor. Thaler’s ownership of the machine also didn’t establish his entitlement to the patents on the basis he had advanced.
His argument included a version of the idea that an owner should own what their property produces. A tree produces fruit; the tree’s owner owns the fruit. But an invention is intangible, and the court didn’t accept that this property principle supplied the missing route to patent rights.
That might sound painfully technical until you imagine trying to use the same reasoning everywhere. Who owns a result produced using leased computing power, a licensed model, a customer’s data, and a researcher’s carefully designed experiment? The tree analogy starts looking a little under-watered.
The court also emphasized the limits of the case. It wasn’t deciding the general patentability of inventions made using AI. Nor was it conducting a scientific investigation of whether DABUS genuinely invented these particular things. The legal outcome followed the applications as presented and the statute being applied.
That leaves plenty to discuss. It simply keeps the discussion attached to what the judges actually decided, which is helpful when headlines have already sprinted three blocks ahead.
South Africa’s patent needs its context
South Africa granted a patent in 2021 listing DABUS as inventor and Thaler as owner. That is a real and important event in the story. It also needs a large explanatory asterisk.
South Africa’s system did not conduct the kind of substantive patent examination readers might assume from the word “granted.” The registration therefore shouldn’t be described as a comprehensive judicial endorsement of machine creativity or a scientific determination that DABUS independently conceived the invention.
The Artificial Inventor Project itself acknowledges the distinction between formalities examination and substantive examination. The grant demonstrates that this application passed through that system with DABUS named. It doesn’t mean every other jurisdiction was presented with the same process and inexplicably chose to be difficult.
Different procedures can produce different outcomes before we even reach deeper disagreements about inventorship. That is less cinematic than “one country recognized the machine’s genius,” but it tells us much more about what happened.
Australia’s early opening didn’t survive appeal
Australia briefly supplied the headline that the campaign wanted. In 2021, a Federal Court judge accepted a route that allowed an AI system to be treated as an inventor. In April 2022, the Full Court reversed that decision.
The appeal examined the statutory framework and the relationship between the inventor and whoever is entitled to the patent. Naming DABUS didn’t satisfy the relevant requirement. The first decision remains part of the history, but it isn’t the final Australian position in that litigation.
This is one reason I get suspicious of articles that freeze a legal story at its most exciting afternoon. A trial judgment can become a worldwide headline before an appeal changes the result. Readers then keep encountering the older celebration years later, as if nothing happened afterward.
DABUS deserves a timeline, not a victory montage. There were arguments that found support, arguments that failed, and procedural paths that closed. All of those belong in the story.
AI-generated conceptual illustration above; this is an imagined workspace, not a photograph of DABUS or its actual inventions.
Germany, Switzerland, and why details matter
The German Federal Court of Justice’s June 2024 DABUS decision also required a natural person to be named as inventor. It allowed an adequately clear designation of a person accompanied by a statement that the person had prompted the AI system to generate the invention. That wasn’t recognition of the machine as inventor.
Switzerland added another development in June 2025. Its Federal Administrative Court rejected naming AI as inventor but allowed a path forward with Thaler named, sending the matter back for further handling. This was a procedural and inventorship development, not a declaration that DABUS had acquired legal personhood.
These outcomes resist the tidy “countries for robots versus countries against robots” map. The details include how someone’s contribution is described, what each statute requires, and whether an application can proceed in a particular form.
Even the phrase “AI-generated invention” can hide disagreements. Does it describe a machine proposing the entire technical solution? A researcher making choices among candidates? A long sequence of prompts, constraints, tests, and revisions? Those processes may sound similar in a headline while differing in legally important ways.
The European story continued in 2026
The European Patent Office’s earlier DABUS appeals rejected naming a machine as inventor. A later divisional application tried a different designation: Thaler was named, with additional wording explaining DABUS’s role.
In February 2026, appeal T 0528/25 was dismissed. The decision addressed the inconsistency between naming a person and accompanying statements attributing autonomous conception to DABUS, alongside procedural questions about amended requests.
That is an important update because simply inserting a person’s name wasn’t treated as an automatic fix. The designation and the account of how the invention arose still had to make sense together.
To me, this is the least futuristic and most revealing part of the whole saga. We have extraordinary software, international litigation, philosophical arguments about creativity, and then we end up back at a very ordinary requirement: your explanation needs to agree with the rest of your explanation.
It’s also a reminder that European patent procedure and national patent procedures aren’t interchangeable. A result in Germany doesn’t automatically tell us what the EPO must accept. Calling both “Europe” can erase the exact distinction that explains the outcome.
The American guidance changed after the original article
The USPTO issued revised guidance in November 2025, replacing its February 2024 guidance in full. The central message was that the ordinary inventorship standard applies regardless of AI assistance. AI remains a tool, not a joint inventor.
The revised guidance focuses on conception: the formation of a sufficiently definite idea of the claimed invention. It also clarified that the joint-inventorship analysis used for multiple people shouldn’t be applied as though a lone person and an AI were two legal collaborators.
That matters because an article relying only on the earlier guidance can now describe a superseded approach. The new guidance didn’t make DABUS an inventor. It clarified how existing doctrine should apply when people use AI.
There’s no useful universal percentage here. “The software did 70 percent” sounds satisfyingly measurable until you ask what the percentage represents. Time? Words? Failed prototypes? Computing costs? The legally relevant contribution isn’t necessarily the longest or most expensive part of the work.
Why credit can get messy very quickly
Consider a hypothetical designer trying to improve a reusable food container. She identifies the problem, chooses materials, supplies technical constraints, uses a model to generate possibilities, rejects most of them, and develops a particular workable structure. Describing her as merely the person who pressed a button would erase a lot.
Now imagine someone buying access to a system, accepting its first output, and claiming every creative step as their own. Describing that process in exactly the same way would erase something else.
Between those examples sits a huge amount of actual work. There are people who formulate the problem, people who develop a system, people who generate candidates, people who recognize a promising result, and people who make it function. Some may qualify as inventors under the relevant law. Others may contribute valuable work without meeting that particular test.
That last distinction can feel unfair if we confuse “inventor” with “everyone who mattered.” A research technician can do essential work. A funder can make the project possible. A manager can keep the lights on. Their importance doesn’t automatically settle inventorship.
I see a similar confusion whenever a finished creative project gets reduced to one name. The name can be accurate for one purpose while leaving out an entire network of effort. With AI, the network becomes harder to inspect, and the temptation to tell a simpler story gets stronger.
That’s why documentation matters so much. Preserve the process, the decisions, the revisions, and the contributions while they’re happening. A confident memory reconstructed months later is a flimsy substitute for a record.
Creativity doesn’t require us to pretend a machine has feelings
I’m interested in machine creativity. I’m also interested in using that phrase carefully. We can discuss a system producing an unexpected design without imagining that it felt proud, worried about criticism, or celebrated with a glass of wine afterward.
Surprise is an experience for the observer. It doesn’t, by itself, tell us what happened inside the system. Something can be new to me because I’ve never encountered it, even when it’s already well known somewhere else. Something can be unusual and still be useless. Something can be useful and still be obvious to a specialist.
Those differences become especially important when we move from admiring an image on a screen to restricting who may make or sell a technical product. Patent protection has consequences beyond giving someone a flattering description.
The debate also doesn’t need a contest about whose imagination is purer. People learn from other people, from tools, from accidents, from materials that refuse to behave. We have never created in a vacuum. The interesting question is how to identify contributions when the tools produce results that their users didn’t explicitly specify.
That question is difficult enough. Adding an imaginary emotional life to the computer doesn’t help us answer it.
Who benefits if the law changes?
A machine doesn’t need a plaque on its wall. People and companies, however, have very practical reasons to care who can obtain exclusive rights. Any proposal to change inventorship law should be judged partly by where those rights would go.
Would recognition encourage useful disclosure? Would it support the expensive work of developing a promising result into a product? Would it make records more honest? Those are sensible arguments to examine.
So are the concerns. If a small number of organizations can generate and file enormous numbers of applications, what happens to smaller researchers trying to enter the field? Who pays to examine the filings? Who can afford to challenge a weak patent? Faster production of technical proposals doesn’t automatically mean a healthier system.
I don’t think the answer is obvious. A rule intended to reward creativity can end up rewarding whoever owns the infrastructure. A rule intended to protect individual inventors can become awkward when their tools change. Both possibilities deserve attention without pretending either outcome is inevitable.
We should also ask what public benefit the proposed exclusivity purchases. The point of a patent system isn’t simply to reward anything impressive. It involves a bargain about disclosure, investment, competition, and access. If we change the bargain, we should be able to explain why the public is getting a better deal.
The paperwork can tell the truth without becoming science fiction
One appealing part of the DABUS campaign is its insistence that records should describe what happened. Taking credit for work you didn’t do is a bad foundation for any creative field. I’m entirely comfortable saying that.
But truthful description requires precision in both directions. We shouldn’t minimize a researcher’s actual contribution because a computer was involved. We shouldn’t inflate that contribution because a patent form wants a person’s name. And we shouldn’t inflate a machine’s capabilities because “it dreamed up an invention” makes an excellent opening sentence.
A useful account would explain what the system received, what it produced, what people selected or changed, and what evidence supports the technical result. That account may be less dramatic. It is also much harder to replace with a convenient myth.
There’s room for wonder in that. I find the process fascinating: an idea takes shape through a strange combination of constraints, computation, judgment, and stubborn testing. We don’t have to turn the computer into a lonely artist to appreciate the result.
The question I keep coming back to
DABUS doesn’t resolve the future of creativity. It exposes how many separate questions we hide inside that word. Can a system produce a useful design? Who conceived the claimed invention? Who may own the rights? What should the law reward? What evidence would convince us that the account is accurate?
Those questions overlap, but they don’t collapse into one answer. A beautiful output can leave the ownership problem untouched. A rejected application can leave the technical possibilities wide open. A granted patent can tell us less about a machine’s capabilities than the headline suggests.
I’m definately curious about what computers will help us make. Better materials, less waste, tools that solve problems we’ve tolerated for too long: there is plenty worth wanting. I’m equally interested in whether we can keep an honest record of who contributed and who gets to benefit.
Because if the future arrives carrying a clever little container, I’d like us to look inside the ownership arrangement before we start applauding. The invention may be surprising. The distribution of the rewards doesn’t have to be a mystery.
For an evening of making something with my own hands, a wooden marble-run kit sounds very appealing. Build a section, send a marble through, work out what needs adjusting and get a little too excited when it finally goes where you intended.
Related Reads You Might Enjoy
Your Lifestyle Activates Genes: The Science of Epigenetics
Light Has Weight: The Strange Physics of Photons
The Forgotten Inventions of the 1800s
20 Theoretical Inventions That Could Change the World
Microsoft’s Holographic Teleportation
The Invisible Symphony: How the Universe Flickers Through Our Lives Without Us Knowing
Your Brain Is Not for Sale… Except Maybe It Already Is
The AI That Writes Its Own Rules: Inside DeepMind’s New Era of Algorithmic Creation
Planet Nine Discovery: The Hidden World Beyond Neptune
When Robots Run with Humans: Inside the Beijing Half-Marathon for Androids
The Impossible Signal: Mysterious Radio Pulses Beneath Antarctica
Three Months to Mars: How Starship Is Rewriting the Clock on Interplanetary Travel
Sources and further reading
Updated October 2026.
USPTO revised guidance, November 2025
EPO appeal T 0528/25, February 2026
Australian appeal summary, WIPO
