AI Inventions East Africa: Who Must Patent Law Name?

 The difficult case is not AI as a tool but AI as a source of the inventive idea

A researcher who uses a calculator does not make the calculator an inventor. The same is true when an AI system is used only to search prior literature, rank known compounds or perform routine calculations under human direction. The harder case begins when the system produces a technical solution that the research team did not specifically anticipate: a new molecular structure, engineering configuration, process condition or combination that later becomes the subject of a patent claim.

East African patent law does not yet contain a detailed statutory test for that boundary. That does not mean inventorship can be ignored. Patent systems still require an applicant to identify the person or persons legally responsible for the invention, while ownership may later pass to an employer or company. The practical business problem is therefore evidential: when AI contributed materially, can the applicant still identify a human who actually devised the claimed invention, and can the project records show what that person contributed?

The regional ARIPO route does not create an AI inventor category

Kenya, Rwanda, Tanzania and Uganda are among the contracting states of the ARIPO Harare Protocol. ARIPO can grant patents through a single regional application designating one or more contracting states. Its current patent materials require a description, claims, an abstract and designation of states, and patents remain subject to the ordinary requirements of novelty, non obviousness and industrial applicability. The 2025 amendments to the Harare Protocol and implementing materials do not establish a separate rule saying that an AI system can be named as inventor.

That silence should not be read as permission to invent a convenient answer. ARIPO operates within a framework built around applicants, inventors and legal persons capable of holding rights. WIPO’s current AI policy materials state that qualifying human created inventions remain protectable under existing patent frameworks while debate continues over machine created inventions and over how much human input is required to distinguish the two. In other words, the unresolved policy question is real, but the filing problem must still be handled under existing law today.

Kenya and Uganda show why inventorship and ownership must be separated

Section 30 of Kenya’s Industrial Property Act provides that the right to a patent belongs to the inventor, while allowing assignment and succession. It also addresses unauthorised applications based on essential elements obtained from another person. Uganda’s Industrial Property Act is even more explicit in its definition: an inventor is the person who actually devises the invention. The Act gives the right to a patent to the inventor, requires the inventor to be named in the patent process subject to the statutory declaration mechanism, and contains separate rules for employee and commissioned inventions.

These provisions reveal a point that companies sometimes miss. The person who owns the AI system, pays the cloud bill or employs the research team is not automatically the inventor. The company may ultimately own the patent because of an employment rule or assignment, but ownership follows a legal route from the inventive contribution. A manager should therefore resist the temptation to name a senior executive, principal investigator or model operator simply because someone must appear on the form. If that person did not devise the claimed inventive concept, the record can become vulnerable when ownership, validity or entitlement is later challenged.

Prompting alone should not become a shortcut to inventorship

Human contribution can occur at several stages. A researcher may identify a previously unrecognised technical problem, define constraints that lead to the solution, design a novel workflow, select and combine outputs in a non obvious way, recognise an unexpected technical effect, or modify an AI proposal into the form eventually claimed. Those activities may be highly inventive. By contrast, entering a broad instruction such as “find a better compound” and accepting the first output is a much weaker basis for saying that the user devised the technical solution.

The exact threshold will ultimately depend on the applicable law and the facts of the claim. East African statutes do not presently provide an AI specific contribution percentage. A useful working principle is therefore claim focused: ask what the patent claim says is new and inventive, then identify which human contributed intellectually to that subject matter. The relevant question is not who used the software most often. It is who contributed to the conception of the claimed technical teaching.

International AI cases are persuasive warnings, not East African statutes

The DABUS litigation outside Africa has made the human inventor issue visible. WIPO’s materials record that patent applications naming the DABUS AI system as inventor were rejected in examined jurisdictions including the United Kingdom and before the European Patent Office because the applicable systems required a human or natural person inventor. More recent WIPO materials continue to describe AI inventorship as an active international policy question rather than a settled global rule.

East African applicants should use those cases carefully. A foreign judgment does not rewrite Kenya’s, Uganda’s or the ARIPO legal texts. Its value is practical: it shows the difficulty that arises when an application admits that no human devised the invention yet still seeks a patent under a human centred statutory framework. Until East African legislation or authoritative regional guidance says otherwise, the safer position is that AI may assist the inventive process, but a patent application should be able to identify genuine human inventorship rather than manufacture it after the fact.

The strongest protection may be a contemporaneous invention record

AI assisted research changes what a useful laboratory notebook should contain. Traditional records often show experiments, dates, results and contributors. A modern invention record should also show how the AI system was used. It need not preserve every trivial interaction forever, but it should preserve enough evidence to reconstruct the human contribution to the claimed invention.

For a pharmaceutical or biotechnology project, that may include the research question, the parameters or constraints chosen by the team, relevant prompts or model instructions, important model outputs, reasons particular outputs were rejected or selected, subsequent human modifications, laboratory validation, unexpected results and the point at which the team recognised the claimed technical effect. For engineering or software inventions, equivalent records could show architecture choices, design constraints, model generated alternatives and the human decisions that produced the final claimed solution. Version information matters too: a later model may behave differently from the one actually used.

Evidence should also show who had the right to the patent

Inventorship records solve only one part of the entitlement problem. Once the human inventors are identified, the business must establish ownership. Employment contracts, research collaboration agreements, university policies, consultancy terms and assignments should be checked before filing. Uganda’s Act, for example, contains specific rules for inventions made in employment or commission, while Kenyan law permits assignment of the right to a patent. In multi institution research, an invention can have inventors from several organisations even where the commercial parties expected a single owner.

The patent file should therefore be matched with an entitlement file. That file should contain employment status, relevant IP clauses, assignments, contributor declarations and the collaboration agreement that allocates filing and prosecution responsibility. If the project uses biological materials, confidential data or third party technology, provenance and licence records should be kept separately because ownership of a patent does not erase restrictions attached to inputs.

ARIPO could clarify procedure before legislation is rewritten

The region may eventually need legislation if genuinely autonomous machine invention becomes common. But an immediate statutory overhaul is not the only response. ARIPO and national intellectual property offices could issue examination or practice guidance explaining how applicants should deal with AI assisted inventions under existing law. Guidance could clarify that an AI system is a research tool unless and until the law recognises otherwise, that inventorship must be assessed against the claimed inventive contribution, and that applicants should retain evidence of material human input.

Such guidance would improve predictability without pretending that every difficult philosophical question has been settled. It would also discourage two opposite errors: rejecting ordinary AI assisted inventions merely because a model was used, and accepting nominal human inventors where the records show no real human contribution. WIPO’s own SME guidance notes that if legal systems require a threshold of human contribution, transparency about how AI was used may become increasingly important even though traditional patent law usually focuses on what the invention is rather than how it was conceived.

The practical rule is to document the human before the patent is valuable

For East African researchers and companies, inventorship should be discussed while the research is happening, not when a financing round or patent filing deadline arrives. The team should identify potentially inventive contributions at regular project milestones and confirm them against the claims ultimately drafted. Patent counsel can then distinguish routine technical assistance, model operation and true contribution to the inventive concept.

The practical conclusion is narrow but important. Existing East African and ARIPO patent law can accommodate many AI assisted inventions because using a powerful tool does not by itself remove human inventorship. The risk appears where the claimed solution cannot be traced to any genuine human devising. In that situation, simply naming the model owner, user or project leader may create an entitlement problem rather than solve one. The best protection is not a creative label for the AI. It is a clear record showing what the humans actually invented.

Source note. This article is based on the ARIPO Harare Protocol on Patents, Utility Models and Industrial Designs, including the amendments effective from 1 March 2025, and ARIPO patent filing guidance; Kenya’s Industrial Property Act, especially sections 30 and 31; Uganda’s Industrial Property Act, including its definition of inventor and provisions on the right to patent, naming of inventors, employment inventions and patent application disclosure; WIPO, Frequently Asked Questions: AI and IP Policy; WIPO materials on AI inventions and the DABUS cases; WIPO, Artificial Intelligence (AI) and Inventorship, SCP/37/5 (2025); and WIPO’s guidance for small and medium sized enterprises on creating and inventing with AI.

Suggested citation: 

Ronald Serwanga, “AI Inventions East Africa: Who Must Patent Law Name?” East Africa Legal Insight (5 September 2026).