WIPO puts AI-assisted inventorship, machine-made prior art and obviousness on the patent agenda

The story

The World Intellectual Property Organization is moving the patent debate over artificial intelligence beyond the headline question of whether a machine can be an inventor. A Secretariat report prepared for WIPO's Standing Committee on the Law of Patents maps policy choices around three operational problems: how to identify a legally sufficient human contribution when AI assists the inventive process, whether machine-generated disclosures can become prior art, and how access to routine AI tools should affect the test for inventive step. Member states will discuss the material at the committee's November 2–6 session in Geneva and online.

The document is a menu, not a decision. It does not amend national patent laws, direct patent offices to change examination or establish a WIPO-wide rule. Its significance is that it organizes issues that have often been argued separately into choices policymakers can compare. Independent reporting by MLex on October 30 likewise identified proof, prior art and obviousness as the next practical fronts in the AI-patent debate. A corrected companion study on AI and inventorship, published by WIPO on October 28, supplies the comparative legal background.

On inventorship, the prevailing position remains human-centered. The Secretariat says currently available patent-office guidance applies the same inventorship standard whether or not AI was used, and only natural persons may be named. The harder question is evidentiary: when a model proposes structures, compounds, engineering configurations or experimental paths, which human acts amount to conception rather than merely operating a tool? The report says patent offices could clarify, for example, when prompt engineering or model-related work constitutes a substantial contribution to the inventive concept.

One policy option is to require applicants to label AI's role and distinguish machine-generated elements from human contributions. A less prescriptive alternative is to recommend internal records describing the AI tool, relevant training data, human inputs and how outputs were evaluated or transformed. The report also states the objections: conventional tools do not trigger a separate disclosure regime, extra requirements could burden patent preparation, and examiners may struggle to measure AI's contribution without overstating it. In other words, greater transparency may help prove inventorship while creating a new field of subjective disputes.

Prior art creates a different risk. Patent novelty normally turns on what technical teaching was publicly available before the filing date, not on whether a human wrote it. The European Patent Office's published explanation says an AI-generated document may be prior art if it was made public and is enabling—meaning a skilled person can put the teaching into practice. That distinction matters because generative systems can produce huge volumes of plausible technical text, including hallucinations. A reproducible disclosure may defeat a later claim; an unsupported suggestion should not acquire the same force merely because it is searchable.

WIPO's report finds no jurisdiction-specific reform that gives AI-generated material its own prior-art category. It presents two broad paths: keep applying ordinary publicity and enablement rules, or develop targeted standards for machine-produced disclosures. The first avoids discrimination by origin but could increase search and verification costs as synthetic documents multiply. The second might make quality thresholds clearer, but risks fragmenting the definition of prior art across jurisdictions. Deliberate publication of generated technical descriptions also raises the prospect of cheaply crowding a field and narrowing the patent space available to later researchers.

Inventive step—or non-obviousness—may be the most consequential pressure point. Patent law asks what would have been obvious to a person skilled in the relevant art. If competent practitioners routinely use AI to explore combinations, screen candidates or optimize designs, some work that once required inventive ingenuity may become ordinary. The report says routinely available AI tools should be considered within the conventional analysis and notes that predictable application of a known deep-learning method may fail the test. It also offers a more structured option: administrative guidance with examples covering technical effect, data preprocessing, training-data changes and other field-specific contributions.

The corrected inventorship study revisits the DABUS litigation that tested applications naming an AI system as inventor. WIPO's compilation records the dominant result across the jurisdictions that examined those filings: inventorship requires a natural person with legal capacity and the ability to conceive. The new policy report argues that governments can look beyond that binary. Its options include preserving human-only inventorship, recognizing an AI contribution through a legally accountable proxy, allowing non-human inventors through legislation, or creating a separate right for AI-generated inventions. Those more radical paths would require substantial lawmaking and leave ownership, accountability and incentives unresolved.

INNOVOX analysis: near-term practice is likely to change through documentation and examination guidance before legislatures recognize machine inventors. Companies already record experiments for scientific integrity, regulation and trade-secret management. AI-assisted R&D adds a patent reason to preserve who set the objective, chose the model and inputs, rejected outputs, designed verification and formed the claimed solution. A clean record cannot manufacture inventorship where no human conceived the invention, but it can make a later contribution analysis less dependent on reconstructed memory.

What to watch is whether the committee turns the Secretariat's menu into a defined international workstream. Agreement to collect patent-office examples, study synthetic prior art or develop recordkeeping guidance would influence applicants without changing treaties. Divergence would also be informative: countries may retain human-only inventorship while differing on disclosure duties and the capabilities attributed to the skilled person. The key discipline is to separate agenda-setting from rulemaking. WIPO has framed the decisions; member states and national legal systems still have to make them.

INNOVOX analysis

The immediate issue for research teams is evidence, not machine personhood. If AI contributes to conception, the defensibility of a patent may depend on contemporaneous records showing who framed the technical problem, selected outputs, tested results and formed the claimed inventive concept.

What to watch

Watch the November 2–6 committee session for member-state support for patent-office guidance, applicant recordkeeping or studies on AI-generated prior art. Any consensus work item would matter; discussion alone should not be reported as a new WIPO standard.