03 · What You Need to Know
Why Patent Law Distinguishes Human Inventors From AI Systems
What does it mean to be legally recognized as an inventor?
A patent inventor is a person whose contribution satisfies the applicable legal requirements for inventorship in relation to the claimed invention.
Inventorship is not simply recognition for participating in a research project. Nor does it necessarily belong to the person who owns the equipment, funds the research, supervises the laboratory, or submits the patent application.
In United States patent law, inventorship is closely associated with conception of the claimed invention. The assessment concerns the actual inventive contributions made by natural persons.
Other patent systems use their own statutory language and legal tests, but the distinction between an inventor and a patent applicant or owner remains important.
Inventor
The natural person whose contribution meets the relevant legal standard for inventorship.
Patent applicant or owner
The person or legal entity entitled to apply for or hold patent rights under applicable law and contractual arrangements.
A university may own patent rights without being named as an inventor. Likewise, a researcher may qualify as an inventor while being contractually required to assign the invention to an institution.
The question of ownership of an AI-assisted research invention therefore requires a separate assessment.
What is DABUS, and why did its patent applications matter?
DABUS, an abbreviation for Device for the Autonomous Bootstrapping of Unified Sentience, is an AI system associated with inventor and researcher Stephen Thaler.
Patent applications were filed in several jurisdictions identifying DABUS, rather than a human being, as the inventor of inventions described in the applications.
The applications became important test cases because they asked patent offices and courts to decide whether existing inventorship laws permitted an artificial intelligence system to be named as an inventor.
The dispute was not simply about whether the proposed inventions were technically useful. It concerned whether a nonhuman system could satisfy the legal identity requirement for inventorship.
Different jurisdictions considered the issue under their respective patent statutes, producing a body of decisions that researchers should understand before making claims about AI inventorship.
What did United States courts decide about AI inventorship?
In Thaler v. Vidal, the United States Court of Appeals for the Federal Circuit affirmed that an inventor under the United States Patent Act must be a natural person.
The court interpreted the statutory term "individual" as referring to a human being. Accordingly, an AI system could not be designated as an inventor under the governing legislation.
The decision did not establish that all inventions developed with AI assistance are unpatentable. It addressed whether AI itself could satisfy the statutory requirement for being named as an inventor.
That distinction remains essential.
In November 2025, the United States Patent and Trademark Office issued revised inventorship guidance for AI-assisted inventions. The guidance confirmed that the ordinary inventorship standard applies and that AI systems are treated as tools rather than inventors.
The revised guidance replaced the USPTO's February 2024 guidance. Researchers should therefore avoid relying on superseded descriptions of a special AI-specific inventorship analysis.
What did the European Patent Office decide?
The European Patent Office refused patent applications identifying DABUS as the inventor.
In decisions J 8/20 and J 9/20, the EPO's Legal Board of Appeal upheld the refusals. Under the European Patent Convention, an inventor designated in a European patent application must be a person with legal capacity.
An AI system does not satisfy that requirement.
The decisions also addressed attempts to establish entitlement to the inventions based on ownership of the AI system. Merely owning a machine does not, by itself, establish the legal entitlement claimed in those applications.
The EPO's position does not categorically exclude inventions developed using AI tools. Instead, applicants must comply with the existing inventorship and other patent requirements.
What did the United Kingdom Supreme Court decide?
In December 2023, the United Kingdom Supreme Court delivered its judgment in Thaler v Comptroller-General of Patents, Designs and Trade Marks.
The Court concluded that DABUS could not be an inventor under the Patents Act 1977 because the statutory concept of an inventor requires a natural person.
The Court also rejected the proposition that ownership of the AI system automatically entitled Thaler to apply for patents for inventions attributed to DABUS.
Importantly, the judgment addressed the interpretation of existing legislation. It did not resolve every policy question about whether future legislation should recognize machine inventors.
For researchers, the practical implication is that AI-generated technical output cannot simply be attributed to the AI system as a named inventor in a United Kingdom patent application.
Have all countries reached exactly the same conclusion?
No. Patent law is territorial, and national legal systems may interpret inventorship and entitlement differently.
South Africa became particularly prominent in the DABUS debate when a patent identifying DABUS as an inventor was granted through its patent registration system.
However, that outcome must be interpreted carefully. South Africa's patent system has historically operated without the same substantive pre-grant examination process used by offices such as the USPTO and EPO.
A patent's registration therefore should not automatically be interpreted as a judicial determination that AI qualifies as an inventor under all applicable legal principles.
Other jurisdictions have considered DABUS-related applications through administrative proceedings and litigation, sometimes involving different procedural histories.
Researchers should avoid describing the international position as completely uniform. The more accurate conclusion is that major substantively examined patent systems have rejected naming AI as an inventor under their existing laws, while international debate continues.
Can a human be named as inventor when AI contributed to the invention?
Yes, provided that the human satisfies the applicable inventorship requirements.
The relevant question is not simply whether AI was used. It is whether the named human inventors made the legally qualifying contributions to the claimed invention.
Consider the difference between a researcher who develops a novel technical mechanism and uses AI to explore possible implementations, and a researcher who merely receives an AI-generated solution and performs routine verification.
The two situations may raise different inventorship questions.
Writing prompts, selecting an output, recognizing potential commercial value, or conducting experiments does not automatically establish inventorship. Those activities must be evaluated in the context of the actual inventive contribution.
At the same time, extensive AI assistance does not automatically disqualify a researcher who genuinely contributed to the conception of the claimed invention.
The broader question of patentability of an AI-assisted research invention also involves novelty, nonobviousness or inventive step, subject-matter eligibility, and disclosure requirements.
What if AI independently generates the entire inventive concept?
This is the situation that creates the greatest difficulty under legal systems requiring natural-person inventors.
If a claimed invention was generated without any human contribution satisfying the relevant inventorship standard, identifying a qualifying human inventor may be impossible.
A researcher cannot necessarily resolve the problem by naming the person who owns the AI system, operates the computer, pays for the subscription, or conducts routine testing.
Nor is it appropriate to invent a human contribution merely to satisfy a filing requirement.
The technical question of whether generative AI can produce an inventive research idea must therefore be distinguished from the legal question of who can qualify as an inventor.
A technically promising AI-generated concept may still present unresolved inventorship problems.
Why does incorrect inventorship matter?
Patent applications require accurate information about inventorship. The consequences of errors depend on the jurisdiction, the nature of the error, and whether correction is legally available.
Incorrect inventorship can complicate patent prosecution, ownership arrangements, enforcement, and disputes between researchers or institutions.
In some circumstances, patent systems provide procedures for correcting inventorship. These procedures should not be treated as permission to make unsupported declarations initially.
Researchers should preserve records showing how the invention developed and what each human contributor actually contributed.
Material AI interactions may also be relevant to reconstructing the development process, although retaining prompts does not independently establish legal inventorship.
04 · A Practical Example
An AI System Suggests a New Engineering Mechanism
Hypothetical Example
Identifying the inventor of an AI-assisted device
A university research team is developing a mechanical device that reduces vibration in laboratory equipment.
The researchers use generative AI to explore possible component arrangements. The AI suggests a configuration that appears promising.
One researcher subsequently develops a different mechanical coupling mechanism, establishes how it interacts with the proposed arrangement, and produces a working prototype.
The university considers filing a patent application claiming the resulting device.
Step 1: Identify the claimed technical features
The team distinguishes the AI-suggested configuration from the mechanical coupling mechanism developed by the researcher.
Step 2: Document the human contributions
Research notes, design records, and relevant AI interactions help establish who developed the features eventually included in the patent claims.
Step 3: Apply the relevant inventorship standard
A qualified patent professional assesses whether the researcher's contribution satisfies the legal requirements for inventorship. The conclusion depends on the actual claims and applicable jurisdiction.
Step 4: Distinguish AI assistance from legal inventorship
The AI system is documented as a development tool rather than being named as an inventor in jurisdictions requiring natural-person inventors.
Step 5: Resolve ownership separately
The university reviews employment and intellectual-property agreements to determine who is entitled to apply for and hold the resulting patent rights.
The researcher may qualify as an inventor if the claimed invention includes the legally relevant contribution developed by that researcher. The AI system's involvement does not automatically eliminate that possibility.
If the researcher had merely tested a fully AI-generated solution without contributing to its conception, the inventorship assessment could be different.
