AI agents are now actively participating in drug discovery, chip design, materials research, and software architecture. They produce patentable output at a speed and volume that examination infrastructures are not built to handle. The question of who owns the products of inventions, and whether those inventions are protected at all, is a question that no jurisdiction has fully resolved.
Together, the world’s five largest intellectual property offices account for approximately 85% of patent applications filed worldwide. According to a report from the European Patent Office, five offices – the EPO, Japan Patent Office, Korea Ministry of Intellectual Property, China State Intellectual Property Office, and USPTO – have agreed to strengthen cooperation on AI to improve the quality and efficiency of examinations, and committed to a comprehensive review of the joint AI roadmap first approved in 2021.
You can see the urgency of it by looking at the scale of what they are managing. WIPO revealed that innovators around the world filed 3.7 million patent applications in 2024. This is a 4.9% increase compared to 2023, marking the highest year-on-year growth rate since 2018, with AI and computer technology cited as a central driver. Since the introduction of the Transformer architecture in 2017, the number of generative AI patents has increased by more than 800%, and 2023 was a record year with more than 25% of all generative AI patents filed so far published in a 12-month period, WIPO’s Generative AI Patent Status Report reveals.
The offices considering that roadmap are not discussing a theoretical future. They are already dealing with an ongoing surge in applications, but the underlying examination systems are not keeping up.
AI is a tool, but humans need to think
The USPTO has issued revised guidance providing that AI systems are tools, not inventors, and that only natural persons can be named as inventors in U.S. patent applications that include AI-assisted inventions.
Under the new framework, traditional conceptual standards will apply to all inventions, regardless of the involvement of AI. Conception is the formation in an inventor’s mind of a clear and lasting idea of a complete and working invention. AI systems that suggest new compounds or generate software architectures do not meet that criteria. The humans directing the process must have independently formed the settled ideas.
Advertisement: SCROLL TO CONTINUE
For companies implementing agent AI in R&D, the impact is direct. Any invention powered by AI requires documentation of human contributions at a conceptual level. AI can search, generate, and optimize. Humans must become pregnant. Organizations that cannot clearly draw those boundaries in their documents are filing patents that may not survive examination.
Gap that policy could not fill
The USPTO’s guidance answers questions regarding inventorship. It was left open to others. In a recent analysis, the Oxford Journal of Intellectual Property Law and Practice found that while this framework reaffirms human-only inventorship and treats AI as a tool similar to laboratory equipment or software, it leaves unanswered questions that could result in inconsistent patent trial boards and court rulings and undermine the predictability of the entire system.
Conceptual standards were built for a world where human engineers approach problems from first principles. This is not designed to describe a workflow where an agent processes 10 million combinations of molecules overnight and emerges with three patent-worthy candidates. The person who directed the process made some contribution. Whether the contribution constitutes a legal idea is a question that courts and patent boards have not yet consistently answered.
For all of our coverage of PYMNTS AI, subscribe to our daily subscription AI Newsletter.
