Can patent attorneys use AI? The rule-by-rule map

Yes. No USPTO regulation and no rule of professional conduct prohibits a patent attorney from using AI in prosecution. Every authority that has examined the question since 2024 — the USPTO, the ABA, and the state bars that have spoken — reached the same structure: use is permitted, and your existing duties govern how you use it.

That structure matters more than the permission. The duties that bind you are the ones you already know: confidentiality, candor, verification of what you sign, supervision, communication with clients, and reasonable fees. None of them changed. What changed is what they require of you when a tool drafts, searches, or analyzes.

This page maps each authority to each duty, with verbatim citations, so you can answer the question for your own practice. It was last reviewed on August 3, 2026.

The USPTO issued its guidance on AI use by practitioners on April 11, 2024 (89 FR 25609). Two sentences do most of the work:

"The capabilities of these tools continue to grow, and there is no prohibition against using these computer tools in drafting documents for submission to the USPTO. Nor is there a general obligation to disclose to the USPTO the use of such tools."

So: no ban, and no general disclosure requirement. The guidance then spends most of its length reminding practitioners that existing duties apply with full force.

The central one is the signature rule, 37 C.F.R. § 11.18(b). Presenting any paper to the Office — by signing, filing, submitting, or later advocating — certifies that you made an inquiry reasonable under the circumstances. As the guidance puts it, that inquiry ensures "the paper is not being presented for any improper purpose, the legal contentions are warranted by law, the allegations and other factual contentions have evidentiary support, and the denials of factual contentions are warranted on the evidence." A tool cannot make that certification. A human who reviewed the contents can. That is the operative rule: you may use AI to draft, but a practitioner must verify every filing before signing it.

Two further points from the guidance. First, the duty of candor and good faith (37 C.F.R. § 1.56) is unaffected; where AI use is itself material to patentability (the inventorship context), it must be disclosed. Second, the USPTO flags confidentiality, not permissibility, as the real risk: AI tools may retain what you enter and use it for training or share it with third parties, in breach of a practitioner's obligations under 37 C.F.R. § 11.106. Tool selection is where the ethics analysis actually lives.

The ABA Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512 on July 29, 2024. It is the national baseline, and it is specific.

Competence (Rule 1.1). "To competently use a GAI tool in a client representation, lawyers need not become GAI experts. Rather, lawyers must have a reasonable understanding of the capabilities and limitations of the specific GAI technology that the lawyer might use." This is not static; you must stay current as tools change.

Verification. The opinion is blunt: a lawyer's reliance on or submission of a GAI tool's output "without an appropriate degree of independent verification or review of its output — could violate the duty to provide competent representation as required by Model Rule 1.1." And for tribunals: "output from a GAI tool must be carefully reviewed to ensure that the assertions made to the court are not false."

Confidentiality (Rule 1.6). This is the consent trigger, and its scope is narrower than most summaries suggest. Because many self-learning GAI tools are designed so that their output could lead to disclosure of client information, "a client's informed consent is required prior to inputting information relating to the representation into such a GAI tool." Two boundaries matter. The trigger is self-learning tools — tools that train on your inputs. And the opinion expressly notes that if a lawyer uses a tool "in a manner that does not require inputting information relating to the representation, client informed consent would not be necessary." Where consent is required, it must be real: "merely adding general, boiler-plate provisions to engagement letters purporting to authorize the lawyer to use GAI is not sufficient." Baseline diligence for any tool: read the Terms of Use and privacy policy, and determine whether the tool retains your inputs or asserts rights over them.

Supervision (Rules 5.1 and 5.3). "Managerial lawyers must establish clear policies regarding the law firm's permissible use of GAI, and supervisory lawyers must make reasonable efforts to ensure that the firm's lawyers and nonlawyers comply with their professional obligations when using GAI tools." The frame is the same one you already apply to non-lawyer assistants and outside vendors.

Fees (Rule 1.5). Lawyers who bill hourly "must bill for their actual time." Efficiency gains belong to the client, and "a lawyer may not charge a client to learn about how to use a GAI tool or service that the lawyer will regularly use for clients."

State authorities have followed the same permit-but-govern structure, and one has now written AI into its rules.

Colorado (January 2026). On January 8, 2026, the Colorado Supreme Court adopted AI-specific amendments to the Colorado Rules of Professional Conduct — the first state to amend its ethics rules for AI rather than issue an opinion. A new Scope paragraph [20A] states: "Technology, including artificial intelligence and similar innovations, plays an increasing role in the practice of law, but that role does not diminish a lawyer's responsibilities under these Rules." A revised Comment [8] and new Comment [9] to Rule 1.1 make technological competence explicit and add that reliance on technology "does not diminish the lawyer's duty to exercise independent judgment." Note what Colorado did not do: it did not prohibit anything, and it did not create a disclosure regime.

Texas (February 2025). Opinion 705 of the Professional Ethics Committee permits generative AI use, requires a reasonable understanding of the technology before use, requires reviewing a tool's terms and data-security protections before entering confidential information, and holds that "lawyers are responsible for the work product they submit regardless of who (or what) does the original research and drafting."

Florida (January 2024). Ethics Opinion 24-1 recommends obtaining the client's informed consent before using a third-party generative AI program where use would involve disclosure of confidential information, and notes that the concern is mitigated where the tool does not disclose confidential information to third parties.

The USPTO's Rules of Professional Conduct (37 C.F.R. Part 11) track the ABA Model Rules, so the two frameworks converge on the same five duties:

On confidentiality, the practical question is what you may put into what kind of tool. Three categories:

1. Consumer tools that train on inputs. Under ABA 512 and Florida 24-1, putting information relating to a representation into these requires informed client consent first. Many firms simply prohibit it.
2. Commercial tools contractually barred from training, with defined retention. Here the analysis is the vendor-diligence analysis you already run for cloud document management and e-filing: read the terms, confirm no training and no asserted rights over your data, confirm retention and deletion, confirm breach notification.
3. Modes that make no external transmission at all. Lowest-risk category. Nothing leaves, so the disclosure analysis largely collapses.

On candor, prosecution has a sharper version of the verification duty than litigation does. An office-action response argues about what references disclose. If a tool mischaracterizes a reference and you file it, that is your § 11.18(b) certification, and it is also an argument the examiner can dismantle. Verification is not a compliance step; it is the work.

A recurring objection: if I put an unfiled invention disclosure into an AI tool, have I made a public disclosure that destroys novelty under 35 U.S.C. § 102?

The doctrine says no, provided the disclosure is confidential and contractually bound. Every § 102(a)(1) category turns on public availability. For printed publications, the long-settled standard is public accessibility: a document must be "disseminated or otherwise made available to the extent that persons interested and ordinarily skilled in the subject matter or art, exercising reasonable diligence, can locate it" (MPEP 2128, quoting In re Wyer). For public use, MPEP 2152.02(c) states that public use under AIA § 102(a)(1) "is limited to those uses that are available to the public." The residual "otherwise available to the public" clause is likewise a public-availability test.

A transmission to a service provider that is contractually barred from using or disclosing the content is not disseminated to the interested public and is not accessible to it. This is the same doctrine under which firms already use email, cloud document management, e-filing vendors, and translation services on unfiled applications without creating prior art. We are aware of no authority holding that confidential processing by a contract-bound vendor constitutes a public disclosure under § 102.

Two careful caveats. First, the analysis assumes the contract actually says what you think it says. A consumer tool that trains on your inputs or asserts rights over them presents a different set of facts, and it is a Rule 1.6 problem before it is ever a § 102 question. Second, the AIA on-sale bar is a separate doctrine about commercial sales of the invention; it is not implicated by vendor processing, and nothing here addresses it. Read the terms, then decide.

Sparlo is built for patent prosecution, not general legal work. It was designed against the duty map above.

Confidentiality (Rule 1.6). Sparlo runs on Anthropic's commercial API. Under Anthropic's Commercial Terms, Anthropic may not train models on customer content: nothing you enter trains a model. API inputs and outputs are deleted from Anthropic's systems within 30 days per its retention documentation, and processing is US-only. Application drafting and review of your own drafts run at zero external calls: those modes make no outbound queries at all.

Candor and verification (§ 11.18(b)). Office-action analysis maps each rejection and checks the examiner's characterization of every cited reference against what the reference actually discloses, with pinpoint support, so the verification you must do anyway is faster and starts from the record. You sign; the tool is built to make your review tractable, not to replace it.

Supervision (Rules 5.1/5.3) and communication (Rule 1.4). Every outbound prior-art query is logged per matter. The egress log gives a supervising attorney, or a client, a complete record of what left the building and when. A client consent pack and per-matter AI policy controls let you set, matter by matter, what the tool may do, and give you the documentation ABA 512 says boilerplate cannot supply.

Evaluation without exposure. The trial is self-serve, no demo call, and can be run entirely on public documents; nothing confidential is needed to evaluate it. Worked examples on real public patents are published at sparloip.com/examples. The Attorney plan is $299 per seat per month, expensable by a single attorney.

This page is informational. It is not legal advice, and it is not an ethics opinion. Rules of professional conduct vary by jurisdiction, ethics opinions are advisory, and both change; the authorities above were verified as of August 3, 2026. Consult the current rules and opinions of your own jurisdiction, your firm's general counsel, or your bar's ethics hotline before relying on any statement here for a specific matter.