AI agents are now active participants in drug discovery, chip design, materials research and software architecture. They generate patentable output at speeds and volumes the examination infrastructure was not built to handle. The question of who owns what they produce, and whether those inventions can be protected at all, is one no jurisdiction has fully resolved.
The five largest intellectual property offices in the world collectively account for around 85% of patent applications filed worldwide. European Patent Office reported that the five offices, comprising the EPO, the Japan Patent Office, the Ministry of Intellectual Property of Korea, China’s National Intellectual Property Administration, and the USPTO, have agreed to strengthen cooperation on AI to improve examination quality and efficiency, committing to a comprehensive review of their joint AI roadmap first endorsed in 2021.
The scale of what they are managing explains the urgency. WIPO found that innovators around the world filed 3.7 million patent applications in 2024, a 4.9% increase over 2023 and the fastest year-on-year growth since 2018, with AI and computer technology named as the core drivers. Since the introduction of the transformer architecture in 2017, the number of generative AI patents has increased by over 800%, with 2023 a record year in which over 25% of all generative AI patents ever filed were published in a single twelve-month period, WIPO’s generative AI patent landscape report found.
The offices reviewing that roadmap are not debating a theoretical future. They are managing a filing surge already underway, and the examination systems underneath it have not kept pace.
AI Is a Tool, but a Human Must Conceive
USPTO issued revised guidance establishing that AI systems are tools, not inventors, and that only natural persons can be named as inventors on U.S. patent applications, including those involving AI-assisted inventions.
Under the new framework, the traditional conception standard governs all inventions regardless of AI involvement. Conception is the formation in the mind of the inventor of a definite and permanent idea of the complete and operative invention. An AI system that proposes a novel compound or generates a software architecture does not meet that standard. The human directing the process must have independently formed that settled idea.
Advertisement: Scroll to Continue
For companies deploying agentic AI in research and development, the implication is direct. Every AI-assisted invention needs documented human contribution at the conceptual level. The AI can search, generate and optimize. The human must conceive. Organizations that cannot draw that line clearly in their documentation are filing patents that may not survive review.
The Gaps the Policy Did Not Close
The USPTO guidance answered the inventorship question. It left others open. The Oxford Journal of Intellectual Property Law and Practice found in its recent analysis that while the framework reaffirms human-only inventorship and treats AI as an instrument analogous to laboratory equipment and software, unresolved questions remain that could yield inconsistent Patent Trial and Appeal Board and court decisions, eroding predictability across the system.
The conception standard was built for a world where a human engineer worked through a problem from first principles. It was not designed to describe a workflow where an agent processes ten million molecular combinations overnight and surfaces three candidates worth patenting. The human who directed that process contributed something. Whether that contribution constitutes conception under the law is a question courts and patent boards have not yet answered consistently.
For all PYMNTS AI coverage, subscribe to the daily AI Newsletter.
