AI Cannot Be the Inventor: What the USPTO’s Current Position Means for Patent Practice

AI can help generate alternatives, test designs, draft text, and search technical literature. It still cannot be named as an inventor in a U.S. patent application. The harder question for patent teams is not whether the AI gets a name on the filing—it is whether the human contribution to conception is clear, supportable, and correctly documented.

The USPTO Simplified Its Position in November 2025

In November 2025, the USPTO issued revised guidance for AI-assisted inventions and rescinded its February 2024 guidance in its entirety. The revised position is deliberately simpler: there is no special inventorship test merely because an AI system was used. The same U.S. inventorship law applies to AI-assisted inventions as to any other invention.

The agency also restated the basic statutory point. An inventor must be a natural person. AI systems are tools used by human inventors; however capable the tool becomes, it does not acquire inventor status.

The Real Question Is Human Conception

For a patent team, the practical inquiry therefore moves away from measuring whether the AI was “important” and toward the ordinary question of conception. Who formed the definite and permanent idea of the claimed invention? Which human contributors conceived the subject matter ultimately claimed?

This matters because an AI-assisted workflow can blur authorship. A researcher may provide the problem and constraints; a model may generate dozens of architectures; a second engineer may identify the one that actually works and materially reshape it. The filing team still needs a legally coherent account of the human contribution.

Do Not Turn Every AI-Assisted Invention into a Pannu Exercise

The earlier 2024 guidance generated confusion because practitioners sometimes read its discussion of the Pannu factors as if every human using AI had to prove a special threshold of contribution against the machine. The 2025 revision rejects that framing. Pannu remains relevant to joint inventorship among natural persons; it is not a test for allocating inventorship between a human and an AI system.

That is a useful correction. The AI is not a competing co-inventor. The legal task is to identify the natural person or persons who meet the ordinary inventorship standard.

What Patent Teams Should Preserve

The revised guidance reduces doctrinal complexity, but it does not make recordkeeping irrelevant. In AI-heavy R&D, contemporaneous evidence can help reconstruct who contributed what when inventorship is later questioned.

problem statements and technical constraints supplied by human contributors
key prompts or experiment instructions when they illuminate human conception
model outputs that materially shaped later work
engineering notebooks, commits, prototypes, and test results showing human selection or modification
claim-development notes connecting the final claimed subject matter to identified human contributors

The objective is not to archive every token exchanged with a model. It is to preserve enough reliable evidence to explain the human inventive contribution if the issue becomes material.

Why This Matters for Chinese-Originated U.S. Filings

For Chinese companies, inventorship is often reviewed late—after a Chinese filing, a PCT application, or a rapid internal AI-assisted drafting process has already fixed much of the record. That can create avoidable problems when the U.S. claim set evolves.

A useful U.S. filing review should therefore ask whether the inventors named in the priority application still correspond to the subject matter being claimed, whether AI-assisted development occurred before or after the priority filing, and whether later claim amendments implicate contributors who were not originally identified.

AI Output Evaluation Is Different from Inventorship

There is also a broader lesson for Patent AI. The fact that AI cannot be an inventor does not mean AI is irrelevant to invention or patent work. It means legal accountability remains attached to humans. The same separation appears in patent analysis: a model can propose an answer, but a qualified human must verify the record, the claim scope, and the legal conclusion before relying on it.

Takeaway

The USPTO’s current position is not anti-AI. It is a normalization rule: use AI as a tool, but apply ordinary inventorship law to the humans. For companies building AI into R&D, the practical advantage comes from better workflows for documenting conception—not from trying to quantify whether the machine was “creative enough.”