EP 4 067 251 A1 stems from a divisional application of EP 18 275 163=EP 3 564 144 A1 and relates to a food container.
Brief outline of the procedure
The ED refused the application as none of the requests on file contained a designation of the inventor that met the requirements of Art 81 and Rule 19.
The MR comprised:
- EPO Form 1002 which designates the appellant, i.e. Stephen L. Thaler, as the inventor.
- An Addendum to EPO Form 1002 which contains inter alia statements to the effect that a machine called “DABUS” conceived of the present invention.
AR1 was filed with claims 1-9 and EPO Form 1002, together with an amended page 1 of the description and an amended Addendum to the designation of inventor.
AR2 was filed with claims 1-9 and EPO Form 1002, together with the Addendum to the designation of inventor filed and the description as originally filed.
With its statement of grounds of appeal, the applicant filed a new AR3 comprising only EPO Form 1002 with the following designation of inventor: ” “Stephen L. Thaler, who prompted the artificial intelligence DABUS to generate the invention.”
In response to the board’s communication under Art 15(1) RPBA, the applicant submitted arguments on the merits and filed a new AR4 containing only EPO Form 1002 and designating “Stephen L. Thaler” (the applicant) as the inventor. No Addendum to Form 1002 and no amendment to the description are part of the fourth auxiliary request.
The board confirmed the refusal.
The applicant’s point of view
In light of the decision J 8/20. regarding the parent application EP 18 275 163, the mere fact that the applicant owned, used and trained the DABUS machine is sufficient for him to be designated as inventor under the EPC.
There is no case law which would prevent the applicant as the user and the owner of DABUS from designating himself as inventor under European patent law as was done in EPO Form 1002.
The Addendum to EPO Form 1002 and the note in the description were submitted to ensure that there is no ambiguity for the EPO or the public about what the applicant’s involvement was, and what it was not.
Another important reason for the explanations given in the Addendum and the note in the description is that in many jurisdictions, e.g. under US law, misrepresenting oneself as an inventor is a crime.
Regarding AR1 and 2, the wording in the Addendum has been simplified and shortened, and no longer contains any controversial statements.
The wording used has previously been accepted by the German Federal Court of Justice (BGH) with decision of 11.06.2024, X ZB 5/22, which was not taken into account by the ED.
Contrary to the ED’s opinion, there is no need to resort to assumptions about what involvement of the appellant the word “cause” might suggest or not.
His extent of involvement is explicitly mentioned in the Addendum, beyond any doubt. In light of J 8/20, the fact that the appellant owns and used DABUS is sufficient for him to be designated as inventor.
The designation of inventor according to AR3 comprises exactly the same wording as used in the proceedings before the BGH, X ZB 5/22. This request is filed in conformance with Art 12 RPBA.
The applicant trusted that the wording in the Addendum of AR1-2, “caused […] DABUS to conceive of the invention”, is a correct translation of the German original “DABUS dazu veranlasst hat, die Erfindung zu generieren”, that was accepted by the BGH. AR3 was submitted to remove any potential misunderstandings based on the translation.
A refusal of the present application would be at odds with the substantive reasoning given in the corresponding national proceedings by the BGH, X ZB 5/22, and the Swiss Federal Administrative Court, decision of 26.06.2025, case No. B-2532/2024.
With regard to AR4, the applicant argued that there are exceptional circumstances which justify its admittance under Art13(2) RPBA, as there is no precedent for how the general guidance of J 8/20 should be applied in practice and the admittance of this request is in the interest of legal certainty at large.
The board’s decision
Inventorship and AI
The board reminded that, pursuant to Art 60(1), the right to a European patent belongs to the inventor or his successor in title. Thus, it follows from the designation of the inventor, who is entitled to the right to a European patent.
The examination whether the requirements of Art 81 and R 19 are satisfied is merely a formal assessment. Under R 19(2), the EPO does not verify the accuracy of the designation of the inventor. However, as indicated in J 8/20, Reasons 4.2.3 and 4.3.8, a formal assessment is required as to whether the request for grant, or the separate statement, identifies an inventor within the meaning of the EPC.
As also pointed out in J 8/20. Reasons 4.3.1 ff and J 9/20, Reasons 4.3.1 ff., the designated “inventor” under the EPC must be a person with legal capacity.
Thus, a machine, such as the artificial intelligence machine DABUS in the present case, cannot be an “inventor” within the meaning of the EPC.
The board referred to J 8/20, Reasons 4.3.9; J 9/20, reasons 4.3.9 as well as to the respective national decisions:
- BGH 11.06.2024, X ZB 5/22;
- BPat G, 11.11. 2021, 11 W(pat) 5/21
- BPat. G, 21.062023, 18 W(pat) 28/20
- UK Supreme Court, 20.12, Thaler v Comptroller-General of Patents, Designs and Trade Marks (Respondent) [2023] UKSC 49
- Court of Appeal England and Wales, 21.09. 2021, Thaler v Comptroller-General of Patents Trade Marks and Designs [2021] EWCA Civ 1374
- High Court England and Wales, 21.09.2020, Thaler v Comptroller-General of Patents, Designs and Trade Marks [2020] EWHC 2412 (Pat)
- High Court of New Zealand, 17.03.2023, Thaler v Commissioner of Patents [2023] NZHC 554
- US Court of Appeal of the Federal Circuit, 05.08.2022, Thaler v Vidal CAFC 2021-2347
- Seoul High Court, 16.05.2024, case No.2023Nu52088
- Federal Court of Australia, 13.04.2022, Commissioner of Patents v Thaler [2022] FCAFC 62
- Swiss Federal Administrative Court (BVerwG), 26.06.2025, case No. B-2532/2024.
For the board, these considerations do not necessarily mean that, if an invention is developed using a machine with AI, there is no legally capable person who can be designated as the inventor. The decision J 8/20 already argues in this direction, Reasons 4.6.2, or J 9/20, Reasons 4.6.2.
It must be considered that patent protection for an invention that was essentially developed using AI would be impossible under the EPC, if natural persons were excluded from being regarded as inventors of AI-generated inventions, since, as indicated above, the EPC requires that a legally capable person is named as the inventor.
However, the provisions on patentability under Art 52 ff. do not rule out patent protection for AI-generated inventions. An invention involves an inventive step under Art 56 EPC if it is not obvious to a person skilled in the art in view of the prior art, which may also be the case for inventions developed by using artificial intelligence (AI).
If an invention is developed using an AI-machine, this also requires some human input or influence, so that it is in principle not excluded to identify one or more natural persons to whom a causal contribution to the invention is attributable, and who can therefore be regarded as inventors within the meaning of the EPC.
The board noted that those considerations were in line with the conclusions of the BGH, X ZB 5/22, on the basis of the German Patent Act, which also requires that an inventor is a natural person.
The BGH found that the designation of a natural person as inventor was also possible if a system with artificial intelligence had been used to find the claimed technical teaching.
The Swiss Federal Administrative Court, case No. B-2532/2024, approved the designation of the appellant, Mr Stephen Thaler, as inventor by concluding that the appellant had sufficient influence on DABUS overall to be considered the inventor, as he was at least involved in providing data and training DABUS, accepted the finished solutions and recognised that they constituted an invention eligible for protection
A human contribution that had significantly influenced the overall success was sufficient, even if the technical teaching had been discovered with the help of an AI system.
The type or intensity of the human contribution was not of decisive importance. The fact that an AI system had made a substantial contribution to the discovery of a technical teaching did not contradict the assumption that there was at least one natural person who was to be regarded as the inventor on the basis of its contribution, so that it was possible and reasonable for the applicant to name at least one inventor even if, in his view, an AI had made the main contribution
The board thus concluded that, in principle, the designation of an inventor under the EPC is also possible for inventions developed using AI.
The MR
For the board, the declarations in EPO Form 1002 on the one hand and in the Addendum on the other hand are not clear and consistent, but rather contradictory. In the context of the formal examination under Art 81 and R 19, it cannot be assessed who is in fact being designated as the inventor, the appellant himself, or (also) the AI system DABUS.
AR1-2
The board shares the ED’s view that these requests contain a similar contradiction as found in the MR.
AR3
Irrespective of the question of admittance under Art 12(4) and (6) RPBA, and notwithstanding the issue whether the use of the German language in EPO Form 1002 is permissible under Art 14(2), (3), (6) and R 3(2), this request is not allowable under Art 81 and R 19.
The board noted that the present case differed from the factual situation underlying the BGH decision X ZB 5/22, in which a corresponding wording on the form was accepted.
AR4
AR4 was not admitted under Art 13(2) RPBA.
Comments
Dr Thaler’s steadfastness is truly impressive.
Whatever he tried, he could not achieve his aim to have an AI designated as an inventor or as co-inventor.
An applicant as the user and the owner of an AI system can designate himself as inventor under the condition that the requirements of Art 81 and R 19 are met and the respective declarations do not contain contradictory statements.
The EPC merely requires that a legally capable, natural, person is named as the inventor. This person can receive the help of an AI, but the AI cannot be designated as inventor.
Inventions using AI are eligible for patent protection under the normal requirements of patentability, especially Art 52-57, 83, 84 and 123(2).
It is interesting to note that the present decision has been taken by the ED in a composition of 4 members and hence, the BA in a composition of 5 members.
The search opinion was rather negative, as a lack of N and IS were major objections in view of the prior art mentioned in the ESR. The merits of the case were never discussed.
Comments
11 replies on “T 0528/25-DABUS again-An AI cannot even be a co-inventor”
Thank you for reporting on this latest development.
I always wondered why Mr. Thaler was so willing to spend so much money (presumably his own, or at least I hope so!) on trying to get an AI designated as an inventor on his patent applications. At least the legal community can be grateful to him for bringing his cases in so many jurisductions: they effectively served as test cases, and clarified the applicable law to some extent.
A few years ago, I saw some people calling for amendments to patent law, which would allow to designate AIs as inventors. Now that the cases have been decided, I don’t see those calls being repeated so often.
Which is probably for the best. Not only there are obvious problems with giving the rights of an inventor to a thing without legal capacity, but I view this as unnecessary.
In his AR3, Mr. Thaler admits that he prompted the AI to generate the invention. And he didn’t only prompt; he also viewed the output, recognized it solved some sort of technical problem, and decided it was worth filing a patent application. In my view, this is sufficient to be an inventor. If it is not, then serendipitously discovered inventions cannot have an inventor either.
@ Extraneous attorney,
Thanks for your comment.
I asked myself the question you asked yourself. Mr Thaler seems to have had enough money to spend it in jurisdictions worldwide. That an AI can be of help when generating an invention, is not at stake.
An AI does not, at least for the time being, work completely autonomously, either it is programmed for certain tasks, or it is prompted. In both cases, a natural person is behind this action. So why insist that the AI is designated as the inventor?
In the present case, the AI clearly outputted a device/product and it was thus irrelevant how the AI came to this result. In view of the documents mentioned in the ESR and the content of the ESO, I would have given little chance for a patent to be granted. Mr Thaler’s motives can thus be seen as provoking a discussion on AI as inventor. One wonders why?
In other cases, merely referring to an AI as part of the invention can lead to a lack of sufficiency, cf. T 48/24, T 1669/2 or T 0161/18. When an AI is part of an invention, sufficiency depends on the training data and on the structure of the AI itself. Data being important assets, I doubt that both training data and the structure, will be happily published.
There is also another problem, can a product/device outputted by an AI, i.e. a machine, can at all be inventive. Legally qualified persons and/or academia have here a vast playground to exercise. I think here at screening and developing new molecules in chemistry. In theory the molecules can be defined by AI, but at the end of the day they should be made available and not just be theoretical.
Where I have also reservations, is when an AI outputs what can be held as general knowledge. It all depends on the prompts and the date of the data used to compute the output.
As far as AI used as help in every days IP work, I noted the rather critical approach of all these tools in IPKat: Is AI software for IP just expensive wrapping paper?
Emerging now are research results on the role of AI in diagnosing and treating illness. These reveal the huge potential of AI to act as assistant to a human doctor but its inability to perceive what a real-life human patient needs, to make them well again.
Likewise, AI’s are proving to be an enormous help to an IP attorney, in summarising the relevant literature, sorting out the relevant case law and even writing the detailed description part of the specification of a patent application.
Comparing a patent attorney with a doctor, I would venture to suggest that an AI is not yet in a position to decide what is the minimal combination of technical features that defines the invention, the subject of the patent application. Once the human inventor and human patent attorney are agreed upon the claim 1 feature combination, they can hand over to their assistant, an AI, the job of filling out the patent application. That is not enough to make the AI a co-inventor though, is it?
The danger is that if you ask an AI to write claim 1 for you, it will be only too pleased to oblige, and then all to easy for you (or your employer) to decree that it was the AI that did the inventing, and not you when you framed the task for the AI to perform.
Do you not think that Thaler, amongst others, is to be thanked for forcing attention on what an AI can do and what it is not yet capable of doing. Compare qualification exams, thheoretical and practical. Already we know that AI’s can pass with ease written exam papers which would-be doctors and vehicle drivers must pass but that they are grotesquely incapable of passing the practical parts of such qualifying exams.
Dear Max Drei,
Whilst I am convinced that AI’s can be an enormous help to an IP attorney, in summarising the relevant literature, sorting out the relevant case law, I am not at all convinced that it can write a detailed description of the invention. It can certainly help writing the introductory part of the description relating to the prior art known to the inventor/applicant, but nothing more.
Each invention is in principle unique, see the logo of the EPO and you would like an AI to write the description. As an AI can in principle only regurgitate what it has been trained to do, I fail to understand how an AI can describe an invention, and less the claims.
I would therefore distinguish between two types of AI, that helping the IP attorney to draft part of the disclosure, and an AI helping to go beyond the known prior art. Whether the second AI can be considered as co-inventor or not, is actually irrelevant. What matters is what comes out at the end, which is different from existing prior art. In this respect, I would think that patent attorneys still have a future, and this for a long time.
Thaler wanted a discussion on whether an AI can be an inventor or at least a co-inventor. It is clear that an AI cannot be an inventor, and hence not even a co-inventor. That an AI can help in some parts of the IP attorney’s work is not at stake, but the finishing touch will remain to be given by a human being.
That an AI can pass an exam is in my opinion due to the fact that the AI can look at past exams and try to see what they all have in common. Exams being mainly based on existing cases, a deep analysis of past exams might help solving new exam cases, but would not be capable to draft a description of something new and certainly not the claims as such.
@ Max Drei, Daniel X. Thomas,
I think most of the topics we are discussing become a lot clearer, when one remembers that an “AI” is actually a huge statistical model that outputs a prediction based on some input.
In fact, you can find many patent applications that are obviously about an “AI”, but where the claims recite a “statistical model”, a “machine learning model” or something similar.
With that in mind, it becomes obvious that an “AI” is a tool and should be treated as such. It would be a mistake to think that it invents, decides, or can do so. In fact, some people have said that “AI does not exist” or “AI is artificial but not intelligent”. I agree with them.
As to AI tools for patent drafting, I think they can have value, with one big caveat: they need input to be useful, and the more the better.
If the client/inventor provides a highly detailed invention disclosure, with plenty of text, drawings, and references to the prior art, all well and good. (The client/inventor may even have enriched their invention disclosure with an in-house AI tool before sending it to the patent attorney, but that is another story.)
But how often does/will that happen, really? I think most patent attorneys can remember cases where they drafted a patent application based on an invention disclosure with only a few sentences, a couple of barely legible drawings and a quick meeting with the inventor to understand what the invention is all about. In mechanical fields, it sometimes happens that the entire invention disclosure is just a CAD file depicting a new product and an explanation about what the product is meant to do. For cases like this, the patent attorney will still be in the picture: presumably their role will shift to preparing input for the AI tool and then iterating on the output. New tools, new practices, same job. Like when patent attorneys switched from paper to computers (a time I am not experienced enough to remember).
Of course, all I am saying could be wishful thinking, full of bias because I would like to continue working as a patent attorney. But perhaps it isn’t either.
@Extraneous Attorney
As to your question regarding Mr Thaler’s legal expenses, there is an answer from the Artificial Inventor Project, a group of lawyers led by Mr Ryan Abbott, a British law professor : the lawyers work pro bono. In his publications, Mr Abbott makes no secret of their objective : the DABUS cases are intended to support patentability of AI-generated inventions without human intervention, which would undermine a cornerstone of patent law focused on human creativity. This extends to copyright law as we can see.
It makes sense to ask another question : cui bono ?
There are huge implications incl. geopolitical considerations (keep the US ahead of China) related to the agenda of the US government which vows to entirely deregulate the AI industry. The removal of of MS Perlmutter, Head of the US Copyright Office, has been an illustration. Some major players in this industry have even called for the deletion of all IP law. See https://ip.com/blog/delete-all-ip-law-why-jack-dorsey-and-elon-musk-just-lit-a-match-under-the-patent-system/
I quote the first two paragraphs of the post, which are telling.
« Jack Dorsey didn’t mince words. In a four-word tweet—“delete all IP law”—he torched conventional thinking about innovation and intellectual property. Elon Musk chimed in with a simple “I agree.” And just like that, two of tech’s most prominent disruptors opened a Pandora’s box: Is it time to burn down the IP system?
These aren’t your average armchair critics. Between them, Dorsey and Musk have built and led companies rooted in proprietary tech. For them to question the very system that helped build their empires isn’t just bold—it’s a direct challenge to the foundations of modern innovation. »
@ Francis Hagel
I can understand why lawyers would want to work pro bono: raise novel legal arguments, take a test case to its conclusion, raise their own profile in the process. The rest is beyond my pay grade, I’m afraid. But this is an important piece of context indeed.
@ Extraneous attorney, Max Drei, Mr Hagel,
It is manifest that there are huge implications including geopolitical considerations, when it comes to AI. If AI could invent, it would make discussions about IS obsolete and the whole patent business could go down the drain. We are however far from this point, even if absolute deregulation seems an attractive possibility for some people.
In this respect, the comment of Extraneous attorney “that an “AI” is actually a huge statistical model that outputs a prediction based on some input” is highly relevant. An AI can only output what it has been trained for, and, not to forget, the output also depends the prompt it has been fed with.
In this respect, I am of the opinion that an AI cannot be an inventor, but it can be a tool. If an invention uses an AI as a tool, sufficiency plays a determining role. In the case of the can, it is possible to obtain the device. This might not always be so easy.
It might for instance be possible to theoretically devise new compounds with the help of an AI, but then it has also to be disclosed, how such a compound can be synthesised. It is not enough to find a compound as such, it has to be obtainable by the skilled person on the basis of the original disclosure. Inventions based on AI have a basic problem of sufficiency.
The use of an AI in the daily work of a patent attorney is a different matter. When reading the contribution of T. Bausch on Kluwer Patent blog, or the various posts of Mrs Rose Hughes on IPKat, one realises that there are hosts of problems to be faced when using an AI, the first one being confidentiality. Analysing case law, analysing a patent or non-patent literature might be a help, but I cannot see the job of a patent attorney and of an examiner taken over by an AI.
Determining the common general knowledge of the skilled person with an AI, can lead to problems, as the date at which some information fed into an AI has to be known. The output of the AI is only becoming state of the art on the day of its publication.
That some lawyers or academics want to test how far they can go with an AI is not a problem as such. That some lawyers work pro bono neither. It is certainly a playground for those people. Whether they will be successful in supporting patentability of AI-generated inventions without human intervention is a different matter.
All in all, be it as inventor, or as a patent attorney, grey cells will still be required.
@ Daniel X. Thomas
I agree. To your last point, I would add this: grey cells will still be required, but I am quite hopeful (if not confident) that they will still be *wanted*.
For all the talk about AI being able to do this or that, the fact remains that there are many kinds of decisions we (collectively, as a society) do not want to be made by a computer, AI or no AI. We already complain about customer service being turned over to AI chatbots: what about decisions taken by a patent office? an attorney? a court? a doctor? a hospital? a military officer? a business executive? The list could go on. At some point, I expect pushback that will bring the discourse about AI back to something more grounded.
Ironically, legal work might end up being among the sectors most transformed by AI, because so much revolves around digesting written information and crafting persuasive written language, something LLMs are (supposed to be) good at.
To the list of relevant contributions, I would add Michael Borella’s “Why AI Will Not Take Over the World” over at the Patent Docs blog.
@ Extraneous Attorney,
Thanks for you contribution.
AI is certainly powerful, and can be helpful, but has limitations which cannot be ignored, e.g. training and prompts. To me, AI is plainly artificial, and by no means intelligent in the usual meaning of the word.
At the end of the line an AI might help to weed out the place, but there has to be a human being to evaluate the result. It is important to know how a result can be achieved so as to not be put in front of a fait accompli. .
The contribution of Michael Borella’s “Why AI Will Not Take Over the World” is highly instructive. Her is the link:
https://patentdocs.org/2026/03/02/why-ai-will-not-take-over-the-world/
In the same vein, the older article in Forbes “Why AI Won’t Take Over The World Anytime Soon” is also interesting. Here is the link:
https://www.forbes.com/sites/bernardmarr/2024/05/13/why-ai-wont-take-over-the-world-anytime-soon/
Blindly following the result of an AI can lead to catastrophes, this is why some regulation is needed, contrary to what some people on the other side of the Atlantic might think or say.
In this respect, the attempts of the EU to act in the digital world are laudable, but need to be better communicated. There is no absolute freedom in the real world, why should this be the case in the digital world?
@Extraneous Attorney
On IA-assisted patent drafting, I attended a conference this week during which a patent attorney shared his experience with a specialised commercial tool. He mentioned a lot of frustration. He even described the experience in rather vivid terms as « mentally exhausting ». My take was that such tools can only used to perform tasks you could give to a junior.
Patent drafting is an exercise which requires a myriad of decisions and interactions between the inventor/applicant and the drafter to achieve a suitable outcome : for example, the definition of the technical field (limited or not to the field of use of interest to the applicant), the degree of detail of the description and drawings, the coverage of the supply chain, the dominant market, which may lead to prefer a European or a US claim drafting grammar. I do not see how an IA tool could replace human beings in these respect.
@Mr Thomas
Today, there are IA drug delivery platforms which include an automated synthesis component. See https://www.biopharmatrend.com/news/servier-signs-1b-plus-multi-target-ai-drug-discovery-partnership-with-iktos-1449/
This does not mean the platform can deliver outcomes without human inputs. Human inputs are critical. At the outset, the definition of a target (molecule capable of binding to a specific protein) is provided by the pharma company. Then, the platform relies on proprietary data and expertise of the pharma company, which is responsible for preclinical and clinical trials, and collaboration between the pharma company and the platform provider.