Using AI in practice before the USPTO: disclosure, confidentiality, and review

The one-line version: There is no general requirement to tell the USPTO that you used an AI tool. What the Office’s guidance makes clear is that every existing duty applies exactly as before — your signature still certifies the paper, you still owe candor to the Office, you still must protect client confidences, and you are still responsible for reviewing everything you file. AI does not lower any of those bars.

Is this guidance still in force?

Yes. The USPTO’s guidance on the use of AI-based tools in practice before the Office, published at 89 FR 25609 (April 2024), remains in force. This is worth stating because much of the public commentary on “AI and patents” has moved on to the inventorship guidance that was later rescinded — a different document on a different question. (On that, see the USPTO’s rescinded AI-inventorship guidance.) The practitioner-conduct guidance was not rescinded.

”Do I have to disclose that I used AI?”

Answered precisely: the guidance does not create a blanket duty to disclose the use of AI tools to the USPTO. There is no per-se “AI was used here” statement you must add to a paper simply because a tool helped you draft it.

What the guidance does instead is confirm that your existing obligations are unchanged by AI involvement:

• Your signature is a certification. Papers filed at the Office must be signed, and the signature certifies — to the best of the signer’s knowledge, formed after reasonable inquiry — that the statements are true and the paper is not filed for an improper purpose. That certification is yours whether a tool drafted the first pass or not.
• The duty of candor and good faith to the Office continues to apply. Nothing about using AI narrows it.
• You must review the output. AI can propose; a person must verify. Filing AI output you have not checked — citations, technical assertions, legal statements — is where the exposure is, not the mere fact of using the tool.

So the disclosure question is largely the wrong question. The operative questions are did you review it and is it accurate and candid — and there the answer must be yes.

Confidentiality: know where your inputs go

The other duty the guidance foregrounds is protecting client confidential information. Before putting a client’s disclosures, draft claims, or prosecution strategy into an AI tool, you need to understand:

• Where the input goes and who can access it;
• Whether it is retained or used to train a model;
• Whether the tool could expose privileged or confidential material to third parties.

Confidential and privileged material does not stop being confidential because it is convenient to paste into a chat box.

The guidance also touches export / foreign-filing considerations and USPTO account-access practices when tools route data or credentials outside your control.

Beyond the USPTO: bar-association guidance

For the broader professional-responsibility framework around competent, confidential AI use, three ethics opinions are useful reference points:

• ABA Formal Opinion 512
• D.C. Bar Opinion 388
• Texas Opinion 705

These address, in general terms, the duties of competence, confidentiality, communication with clients, reasonable fees, and supervision as they apply to generative-AI tools. They are professional-responsibility guidance, not USPTO rules, but they inform the same underlying duties.

The pattern that satisfies all of it

Put simply: let AI draft, but keep a person on the certification. Review every output, protect the client’s confidences, and stay candid with the Office. Do that, and the “must I disclose AI?” question mostly answers itself — you are responsible for the paper either way.