Uspto Resets AI Inventorship, Humans Must Own Claims

The news, not the hype:

The U.S. Patent and Trademark Office (USPTO) published revised inventorship guidance on November 26, 2025, replacing the February 2024 framework. Bottom line: there is one inventorship standard for everyone, AI is treated as a tool, and only natural persons can be inventors. The Office also explained how priority works when foreign applications list AI as an "inventor." See the Federal Register guidance: Federal Register (USPTO guidance PDF).

What changed, precisely:

• Pannu factors clarified: Pannu v. Iolab (Federal Circuit, 1998) is back in its lane. Use the Pannu factors only to decide whether multiple humans are joint inventors. The test is not for "AI + one human" situations because AI is not a person. The three-part Pannu test asks whether each named human (1) significantly contributed to conception or reduction to practice, (2) made a contribution that is not insignificant in quality compared to the whole invention, and (3) did more than recite well-known techniques.

• Priority defined: Priority means the right to use an earlier filing date. The USPTO now says a U.S. filing can claim benefit or priority only if the earlier application names the same inventor or at least one joint inventor in common - and that shared inventor must be a human. If a foreign application lists only an AI, you cannot claim U.S. priority.

• Europe / PCT mechanics: Europe ties priority to applicant identity and ownership under the European Patent Convention (EPC Article 87). To keep both U.S. and EPO/PCT (Patent Cooperation Treaty) paths workable, make sure the priority filing lists at least one human inventor or a human applicant. Joint PCT filings can preserve entitlement when at least one joint applicant is the earlier applicant or a successor in title. See the European Patent Office guidance: European Patent Office (EPC guidelines).

Antibody founders, read this twice:

• Written description vs enablement: Written description shows you actually possessed the claimed invention (for example, representative sequences or structural motifs). Enablement teaches others how to make and use the invention without undue experimentation. Courts have been strict here (Ariad on written description; Amgen v. Sanofi on enablement for broad antibody genus claims). Thin disclosures plus broad functional claims are risky. See the Ariad case overview: Ariad case (Justia).

• Documentation to bank now: Keep clear records of who set design constraints or AI parameters, why certain AI outputs were picked or rejected, code and parameter snapshots, and the human-designed experiments that validated claim elements. Examiners and litigants will ask for this evidence.

Mini-example:

If Scientist A tunes the model, defines selection criteria, and chooses sequences to test, that is likely conception. If Scientist B only clicks "run" and forwards outputs, that is probably not conception.

Receipts, because you'll want them:

• Revised USPTO guidance (Nov 26, 2025) replacing the Feb 2024 version: Federal Register (USPTO guidance PDF).

• Original 2024 guidance overview for context: USPTO 2024 guidance overview.

So what:

AI can speed discovery, but patents still rise or fall on human conception. If you miss that, your "AI-discovered" antibody is just a press release plus legal bills.