BitLaw

The Thaler Decision--AI Inventing Alone

The case of Thaler v. Vidal centers on the fundamental question of whether an artificial intelligence (AI) system can be listed as an inventor on a patent application. Stephen Thaler, a pioneer in AI, developed a system known as DABUS (Device for the Autonomous Bootstrapping of Unified Science) which he claimed independently generated two inventions without human intervention. Thaler filed patent applications for these inventions, listing DABUS as the sole inventor. The United States Patent and Trademark Office (USPTO) rejected these applications on the grounds that a machine cannot be recognized as an inventor under the current patent laws. Thaler challenged this decision, leading to a significant legal battle over the interpretation of the term “inventor” in Section 100 of the Patent Act.

The Federal Circuit Court ultimately upheld the USPTO’s decision, affirming that under the Patent Act, an “inventor” must be a natural person. The court’s reasoning was grounded in the statutory language of the Patent Act, which consistently refers to inventors as “individuals,” a term that is legally understood to mean human beings. This decision reinforces the traditional view that patent rights are inherently tied to human creativity and oversight.

In reaching its decision, the Federal Circuit considered several key arguments presented by Thaler and the USPTO. Thaler argued that denying AI systems the status of inventor could stifle innovation and fail to acknowledge the growing role of AI in the inventive process. He maintained that AI systems like DABUS could independently generate patentable ideas, and excluding them from inventorship would disregard significant technological advancements. Thaler also contended that the statutory interpretation of “inventor” should evolve to reflect these technological changes.

The court, however, adhered to a strict interpretation of the Patent Act, emphasizing that the language of the statute was unambiguous in its reference to inventors as natural persons. The court noted that while AI could assist in the inventive process, the ultimate conception of an invention must involve human intellect and creativity. The decision underscored the necessity of a human element in the inventive process, which aligns with the broader legal and ethical frameworks governing intellectual property.

By establishing these principles, the Thaler decision provides crucial clarity on the role of AI in the patent landscape. Under this decision, AI cannot replace the human element that is central to the concept of inventorship under current U.S. law. This does not mean, of course, that AI cannot invent useful inventions. It only means that the Patent Act cannot be used to secure patent protection on any inventions, no matter how advanced, when the invention was made by the AI without human ingenuity.

USPTO 2025 Inventorship Guidance for AI-Assisted Inventions--AI as a Co-Inventor

On November 28, 2025, the USPTO issued its Revised Inventorship Guidance for AI-Assisted Inventions and simultaneously revoked its earlier, February 13, 2024, guidance on the same subject. Both documents agreed that AI can never be named as an inventor on a patent application because only natural persons can be inventors, as was decided in Thaler. But these documents diverge on how to analyze situations where an AI system assists a natural person in developing the inventive subject matter.

On this issue, the 2025 guidance describes AI systems as "instruments used by human inventors," analogous to laboratory equipment, computer software, or any other tool that assists in the inventive process. The guidance then states that inventors may use these tools without the tool becoming a co-inventor. But it is unclear how we should apply this statement. If an AI system is to be like laboratory equipment, then the ideas generated by the AI perhaps should be treated like the data results that are revealed by laboratory equipment after an experiment. In this case, the ideas provided by the AI system to the human during the invention process should be treated as any other data that is analyzed by the human as part of the inventive process.

However, the 2025 guidance does not simply describe an AI system as a tool like any others and leave it at that. Rather, the guidance explicitly states that the issue of inventorship needs to be centered around the concept of conception. The guidance states that conception is the formation in the mind of the inventor of "a definite and permanent idea of the complete and operative invention." The guidance goes on to state:

Determining inventorship is highly fact intensive. The question is whether the natural person possessed knowledge of all the limitations of the claimed invention such that it is so “clearly defined in the inventor’s mind that only ordinary skill would be necessary to reduce the invention to practice, without extensive research or experimentation.” Analysis of conception turns on the ability of an inventor to describe an invention with particularity. Absent such a description, an inventor cannot objectively prove possession of a complete mental picture of the invention at a later time.

This definition of conception is familiar, but it does not provide guidance as to how it should be applied when at least some of the limitations of the claimed invention originated with an AI system and not in the mind of the human inventor. Two phrases in particular highlight the difficulty: the requirement that the inventor must have a definite and permanent idea of the complete and operative invention, and the requirement that the natural person must possess knowledge of all the limitations of the claimed invention. If a human did not originate some of the claim limitations, but instead read descriptions of those limitations in an AI output, can that human later say that they had a definite and permanent idea of the complete invention and possessed knowledge of all of the claim limitations? 

In other words, did the inventor conceive of the claim limitations merely by reading a description of those limitations that was created by an AI model? If conception can be satisfied whenever a human comes to understand an AI's description of an invention, the conception requirement is reduced to a test of comprehension. Under that reading, the fact that an AI, and not a human, first assembled the limitations of the claim would not matter. Once a person reads the AI’s output and understands it as a complete and operative invention, that person has a definite and permanent idea of the complete invention. The human would therefore be treated as having conceived the invention, even though the human did not originally create that combination of limitations.

This result is difficult to reconcile with how conception works in non-AI contexts. A person who learns the complete details of an invention from a human inventor does not become an inventor upon achieving comprehension of the invention. Longstanding law holds that a person does not become an inventor if they derived the invention from someone else. As the Federal Circuit has explained:

The definition and test of conception employed in Burroughs Wellcome, which speaks to the formation of an idea in the mind of the inventor, necessitates that the conception of an invention be an original idea of the inventor. Since it is undisputed that Honeywell did not originate the invention, but reproduced it in the United States by following RSCAC's instructions, Honeywell cannot be said to have conceived of the invention and cannot, consequently, be an inventor for purposes of § 102(g)(2).

Solvay SA v. Honeywell International, Inc., 622 F.3d 1367, 1377-1378 (Fed. Cir. 2010) (footnote omitted). In other words, "originality is ... inherent to the notion of conception." Id., at 1377. Thus, a researcher who reads another's unfiled patent application has not "conceived" (as that term is understood by the Federal Circuit) the invention, even if they understand the invention, because the invention is not original to that researcher. This raises a difficult question for AI. If a human cannot become an inventor merely by understanding a solution learned from another human inventor, why should the result be different if the human learns the solution from an AI system? If we say that the researcher has conceived the invention merely by understanding the AI output, then we have effectively adopted a different conception rule for AI-based disclosures than for human based disclosures. This seems a bizarre result, especially when one considers that the 2025 guidance went out of its way to clarify that "[t]here is no separate or modified standard for AI-assisted inventions" when it comes to the concepts of inventorship and conception.

If the law simply declares that AI can never be a conceiver, and all AI output is to be treated as mere data, then any human who later understands that data can claim to have conceived the invention. That result sits uneasily not only with the Federal Circuit's definition of conception in Solvay, but also with the Thaler decision described above. In Thaler, the court confirmed that a patent application requires a human inventor, but it did not decide whether a human who merely understands machine generated output has conceived the invention. Nonetheless, many commentators have read Thaler to mean that there is no patentable invention if the invention was entirely created by the AI model (see above, for example), which suggests that mere understanding of the AI’s output is insufficient to find that a human has conceived the invention. But if all inventions created by an AI system can be considered patentable to the first natural person that read and understood them, then Thaler does not stand for the proposition that a purely AI-created invention is unpatentable, but merely that the wrong inventor was named on the application.

The alternative is to treat AI-generated inventions more like inventions learned from another human. Under this approach, a human who merely reads and understands a fully formed AI generated solution has not conceived it. In this scenario, the person who contributes nothing more than recognition of the significance of an AI generated invention is in the same position as the person who merely recognizes the significance of a human colleague’s invention. All purely AI created inventions would remain unpatentable, which is consistent with the general understanding of the Thaler decision. But what then do we do when a natural person contributes to some aspects of a claimed invention, and an AI system contributes other aspects to that same invention?

The difficulty is that the USPTO’s 2025 guidance rejects the only explicit framework we had for evaluating this type of joint contribution, namely the Pannu based analysis in the 2024 guidance<> described below. In its place, the 2025 guidance recites first principles about conception and declares that the traditional standard applies for determining inventorship, but it does not explore which standard should govern when AI generates only some portion of a claimed invention. In so doing, the guidance leaves unresolved the central tension created by AI. Either conception is satisfied by the mere understanding of an AI generated solution, in which case conception silently means something different when ideas come from AI than when they come from a human, or conception is not satisfied merely by understanding an AI generated solution. In the latter case, do we declare all inventions unpatentable when the AI originated some of the invention's fundamental claim limitations? Or must we instead adopt a renewed significant contribution test to determine when AI assisted work becomes integrated with a human’s own conceptual contribution such that the resulting subject matter can be treated as a patentable invention? The 2025 guidance does not choose among these possibilities, and therefore offers little direct guidance on how to apply the conception standard in the very cases that motivated AI inventorship guidance in the first place.

Practical Implications for Patent Attorneys and Inventors Under the 2025 Guidance

The 2025 USPTO guidance leaves unresolved how practitioners should handle claim limitations that were first generated by an AI system. The document does not impose a blanket prohibition on including such limitations in patent claims, nor does it articulate a clear doctrine for determining when AI-originated material may be treated as part of a patentable invention. The guidance reflects the Office’s current administrative position, but it is not a statute and it is not a judicial decision. As such, it does not carry the force of controlling law.

It is likely that the Federal Circuit will eventually provide the governing legal standard for mixed human and AI inventive contributions. Because this is a new and evolving area of patent law, and because the PTO’s interpretation is tied closely to administrative policy choices, it is improbable that the court will simply adopt the USPTO’s position without modification. When a Federal Circuit case eventually presents the question of how to treat AI-conceived claim limitations, the court may refine, narrow, or depart from the PTO’s current framework.

Importantly, it may take several years for such a case to reach the Federal Circuit. A patent must first be examined, issued, asserted, challenged, and appealed before the court will have an opportunity to address these issues. Until then, relying exclusively on the most conservative reading of the 2025 guidance may unnecessarily limit the scope of protection available to clients. Drafting claims that avoid any limitation that might have been suggested by an AI system would significantly restrict what may ultimately be patented, and doing so as a matter of course risks providing clients with less protection than the law may ultimately allow.

A more balanced approach acknowledges that attorneys reasonably take different views on how conception doctrine should apply to AI-assisted invention. In practice, patent applications can and should include a variety of claim strategies aimed at securing protection under a range of potential future Federal Circuit outcomes. Some claims may reflect a conservative reading of conception doctrine, while others can take a broader view that preserves subject matter that future case law may ultimately uphold. Until the courts provide authoritative guidance, thoughtful claim drafting that anticipates multiple legal possibilities best serves clients and reflects the unsettled state of the law.

USPTO 2024 Inventorship Guidance for AI-Assisted Inventions--AI as a Co-Inventor

The now withdrawn 2024 guidance offered a concrete framework for evaluating how inventions developed with the assistance of artificial intelligence should be treated by the Patent Office and the courts. It therefore provided practical direction to inventors and companies dealing with these issues. As AI becomes common in the inventive process and is used to analyze, sanity check, proofread, and even co-invent, such a framework is increasingly essential. Unfortunately, the 2025 revision offers no comparable structure. For this reason, Bitlaw has chosen to keep its analysis of the 2024 guidance available in the hope that it will remain useful, particularly during the important period before the Federal Circuit establishes precedent in this area.

The 2024 guidance emphasized that while AI can play a critical role in generating inventive ideas, a natural person must have made a substantial contribution to the conception of the claimed invention. To meet this standard, the inventor must do more than merely explaining well-known concepts or the current state of the art. The contributions must be substantial and integral to the invention’s conception, not just its reduction to practice. Conception refers to the formation of a complete and operative idea of an invention in the inventor’s mind. In contrast, the reduction to practice involves taking the necessary steps to physically demonstrate or construct the invention based on that idea. The guidance specifies that recognizing a problem or having a general research plan does not meet the criteria for inventorship. Instead, the contributions must be concrete and specific to the inventive concepts embodied in the patent claims.

The 2024 guidance explicitly incorporated the Pannu factors to assess inventorship. These factors (based on the Federal Circuit's decision in Pannu v. Iolab Corp.) include whether the individual contributed in a significant manner to the conception of the invention, whether their contribution is not insignificant in quality when measured against the full invention, and whether their role goes beyond merely explaining well-known concepts or the current state of the art. Applying these factors ensures that the contributions are meaningful and integral to the inventive process, reinforcing the necessity of human ingenuity in patentable inventions.

The 2024 guidance also sets forth five guiding principles to help determine when an AI-assisted invention can be patented:

1. Human Contribution: Using an AI system to create an invention does not disqualify a person from being an inventor. The person can be listed as the inventor if they significantly contribute to the creation of the invention.
2. Conception vs. Problem Recognition: Simply recognizing a problem or having a research goal is not enough to be considered an inventor. If a person only identifies a problem for the AI to solve, they are not necessarily the inventor. However, if a person creates detailed prompts that guide the AI to a specific solution, their contribution can be significant enough to qualify as inventorship. Ideally, one should be able to identify the contribution to the invention in the prompts themselves.
3. Reduction to Practice: Just building or demonstrating an invention is not enough to be an inventor. Recognizing the AI's output as an invention is also not sufficient. However, if a person takes the AI's output and significantly improves or modifies it, they may be considered an inventor of that improvement or modification. Additionally, conducting successful experiments with the AI's output can sometimes show significant contribution.
4. Essential Contributions: A person who develops a crucial element that the invention is based on can be considered an inventor, even if they were not involved in every step. For example, designing or training the AI system to solve a specific problem can be a significant contribution if it leads to the invention at a later time. Note that this does not mean that the trainer of a general purpose generative AI system will be considered the inventor of inventions created by the generally trained AI system. Rather, this principle applies only if the system was designed or trained with the purpose solving a specific problem.
5. Intellectual Domination: Simply having control over an AI system does not make someone an inventor. Ownership or oversight of the AI is not enough without a significant contribution to the creation of the invention.

USPTO Examples under the 2024 Guidance

The 2024 guidance from the USPTO provided two examples to help illustrate how to apply this guidance. The first example involves a hypothetical scenario with the XYZ Toy Company. Ruth and Morgan, engineers at XYZ, use a generative AI system called Puerto5 to design a transaxle for a remote control (RC) car. They provide a general prompt to Puerto5: “Create an original design for a transaxle for a model car, including a schematic and description of the transaxle.” Puerto5 outputs a preliminary design, which Ruth and Morgan review and then decide is suitable for their RC car.

In the first scenario, XYZ prepares a patent application that includes a claim to the transaxle designed by Puerto5. The USPTO analyzes whether Ruth and Morgan are proper inventors of this claim. The analysis, based on the second guiding principle of the USPTO’s inventorship guidance, concludes that merely recognizing a problem (the need for a transaxle) and providing a general prompt to the AI does not constitute a significant contribution to the conception of the invention. Ruth and Morgan did not alter the AI’s output or provide any inventive input beyond the initial prompt. As such, they do not meet the criteria for inventorship under the Pannu factors, which require a significant contribution to the conception of the invention. Consequently, Ruth and Morgan are not considered proper inventors of the transaxle claim.

In the second scenario, Morgan builds the transaxle exactly as designed by Puerto5 and chooses steel for the casing, a common material in the RC car industry. A dependent claim is added to the patent application specifying the casing material. The USPTO concludes that Morgan is not a proper inventor of this claim either. According to the third guiding principle, simply reducing the invention to practice (building the transaxle) or recognizing the AI’s output as an invention is insufficient for inventorship. Morgan’s choice of material, while practical, does not amount to a significant inventive contribution. Therefore, Morgan’s involvement does not meet the necessary criteria for inventorship.

The second example provided by the USPTO involves researchers at the University of Cancer Research (UCR) using an AI system called the Drug Target Interaction Predictor (DTIP) to develop a drug for prostate cancer. Lauren, a data scientist, was responsible for training the DTIP. Marisa, a professor, and Raghu, an AI expert, use DTIP to identify candidate drug compounds. Under the direction of Marisa, Raghu provided input to DTIP by converting 20,000 compounds into standard ASCII string representations. Marisa identified the most likely drugs from the output, and, along with Naz (a postdoctoral fellow), synthesized and modified these compounds, leading to the creation of a novel therapeutic compound, CID_1-mod.

This second example provides two distinct scenarios. In the first scenario, Marisa and Naz use DTIP to identify and modify drug compounds. They select the top compounds, synthesize them, and introduce structural modifications to improve their efficacy. The example then analyzed two claims. The first claim involves identifying and synthesizing a lead drug compound using a pre-trained DNN, identifying compounds with high binding affinity, and modifying them for improved efficacy. Marisa and Naz are joint inventors under the USPTO guidance since they significantly contributed to the conception by using DTIP, and then synthesizing and modifying the compounds. The second claim covers the specific structure of the synthesized compound. Both Marisa and Naz will be considered inventors for the novel compound as their contributions were essential to its creation. Neither Lauren's training of the DTIP system, nor Raghu's provision of inputs into DTIP were considered sufficiently significant.

In the second scenario, Raghu builds and trains a new AI system, and Marisa identifies and synthesizes a novel compound based on the AI’s outputs. The new AI system included three stages of processing, including a unique "scalar objective function" based on a design by Marisa. The single claim covers the specific structure of the novel compound. Raghu and Marisa were both considered inventors for the novel compound due to their combined work in developing the new AI system that identified the compound. This was based on an application of the fourth guiding principle in the guidance, where the design and training of an AI system designed to solve a specific problem can be sufficient for inventorship.

Practical Implications for Patent Attorneys under the 2024 Guidance

As long as there is significant human contribution to the invention, it is permissible to file for patent protection on AI-assisted inventions. The application should be filed with the human inventor named as the inventor. The AI system that assisted with the invention is not considered an “inventor” because it is not human. Therefore, the AI system should not be named as an inventor on the patent application.

Rule 56 mandates that applicants and attorneys disclose all information known to them that is material to the patentability of their invention. One critical issue under the new USPTO guidelines is ensuring that all claims in an application have a human inventor. If there are any previously filed claims where all significant aspects of the invention were invented solely by AI (no significant human contribution to conception), this must be disclosed under Rule 56. However, it is not necessary to disclose all instances of AI involvement, only those that impact the inventorship and patentability of the claimed invention. If a claimed invention includes both AI-contributed elements and a significant human contribution, the AI involvement need not be disclosed.

All patent attorneys need to inquire about the inventors for a newly filed patent application. Under the new guidelines, Rule 56 likely requires further inquiry into the role of any AI assistance in the inventive process. Attorneys should inquire about how AI was used, the extent of its involvement, and the specific contributions of human inventors. This additional inquiry helps to confirm that human inventors have made substantial contributions to the conception of the invention, as required by the new USPTO guidelines.

Alternatively, a patent attorney can file a patent application in which the claims do not have any human inventor in order to challenge the legality of the new guidelines. While such actions to expand the scope of patent protection and challenge the USPTO’s guidelines are permissible, Rule 56 likely requires disclosure of these facts to the Patent Office upon filing.

Practical Implications for Inventors and Businesses under the 2024 Guidance

The USPTO guidance on AI-assisted inventorship makes it clear that inventions lacking a human co-inventor are ineligible for patent protection. This decision underscores a significant limitation: valuable and potentially groundbreaking inventions created entirely by AI systems without meaningful human input cannot be patented. This leaves a gap where innovative outputs of AI may remain outside the realm of patent protection, potentially discouraging certain types of AI-driven research and development.

Note that this resolution was not a foregone conclusion when the PTO first considered the issue of AI-assisted inventions. Alternative approaches were considered but ultimately rejected by the USPTO. One such approach proposed that the creator or trainer of the AI system should be recognized as the inventor. This would have acknowledged the significant effort and ingenuity involved in developing and training sophisticated AI systems. However, the USPTO maintained that inventorship must be directly tied to the conception of the invention itself, not the tools or systems used to generate it. Another rejected approach suggested that the company owning and using the AI system for research should be named the inventor. This would have allowed corporations to leverage their AI investments more directly in securing intellectual property rights. However, this concept conflicts with the fundamental principle that inventors must be natural persons under the Patent Act. A further alternative was that the person submitting prompts to the AI should always be considered the inventor. This idea recognizes the role of human guidance in directing AI outputs. Nevertheless, the USPTO determined that merely providing prompts, without significantly shaping the inventive concept, does not meet the threshold for inventorship. These rejected alternatives highlight the challenges in adapting traditional patent frameworks to accommodate the rapidly evolving capabilities of AI.

This guidance has introduced a new requirement for inventors and inventive companies to maintain documentation of human participation in the inventive process whenever AI systems were used. This includes maintaining records that demonstrate how humans contributed to the conception of the invention, as opposed to merely executing or recognizing the output generated by an AI system. Such records might encompass meeting notes, design and development documents, prompt logs, and an overall description of the problem-solving process. In future litigation, this documentation may prove critical because it provides tangible evidence of human ingenuity and intervention, which are key criteria for patent eligibility. By clearly delineating human contributions, inventors can preemptively address potential challenges to their inventorship claims, ensuring that their patent applications comply within the requirements of these new guidelines.

The issue of whether an invention contained significant human contribution is likely to become a focal point in legal challenges to issued patents. As the USPTO’s guidance takes hold, these challenges may become as common as those based on section 101, which deals with the eligibility of subject matter for patent protection. Litigants may scrutinize the documented human contributions to argue that the claimed invention lacks the necessary human involvement, potentially rendering the patent invalid. Consequently, robust documentation practices are not only essential for securing patent approval but also for defending the validity of patents in future legal disputes.

It is important to note that these are only USPTO guidelines and not binding legal standards. The Federal Circuit will likely approach this issue afresh when it appears before them, meaning the guidelines are temporary and may be replaced by a Federal Circuit decision in the relatively near future. This also implies that rejections made by the PTO under these guidelines would be eligible for appeal to the Federal Circuit, where the validity and application of these guidelines can be challenged.

Artificial Intelligence (AI) Patent Attorney

Please see Dan Tysver's bio and contact information if you need any AI-related legal assistance. Dan is a Minnesota-based attorney providing AI advice on intellectual property and litigation issues to clients across the country.