Key Takeaways

• Using generative-AI tools for patent drafting creates more than efficiency gains. AI-assisted patent drafting can create patentability risks associated with novelty, inventorship, § 112, and later, enforceability.
• Public accessibility, not the use of AI itself, drives the statutory-bar analysis. Disclosing an invention to an AI system may qualify as a “printed publication” or as making it “otherwise available to the public” under 35 U.S.C. § 102(a)(1), thereby creating a statutory bar risk that could preclude patent protection.
• AI can assist inventors and patent counsel, but it cannot be the inventor. Inventorship still requires human conception.
• Companies using AI in invention disclosures or patent drafting should adopt guardrails. Use closed AI environments for all invention disclosure and patent drafting, avoid AI tools accessible to the public for unpublished inventions, preserve human control over conception and drafting, and document the workflow.

Generative-AI tools are increasingly being used across the patent workflow, from internal invention disclosures to draft specifications, claim language, summaries and background sections. The efficiency gains can be real. The legal risks are, too.

For patent applicants and prosecutors, the first question should not be whether AI can help draft an application. It often can. The critical question here is whether disclosing invention details to an AI tool or relying on AI-generated drafting creates patentability or enforceability problems that did not exist before the invention was entered into the system.

The Statutory-Bar Question Turns on Public Availability

35 U.S.C. § 102(a)(1) bars patenting if, before the effective filing date, the claimed invention was patented, described in a printed publication, in public use, on sale or otherwise available to the public. The statute also contains a one-year grace period from public disclosure for certain inventor-originated disclosures. In the AI setting, the core question is not whether the disclosure was made to a machine: it is whether the disclosure became public, publicly accessible or otherwise available outside a confidential relationship.

Established case law creates a framework for public disclosures that can be applied to the AI setting. In the printed-publication context, the Federal Circuit has repeatedly emphasized that the key question is public accessibility. The Federal Circuit has made clear that the printed-publication inquiry turns on public accessibility. In re Klopfenstein, 380 F.3d 1345, 1348, 1350 (Fed. Cir. 2004). In In re Hall, the court held that a single thesis qualified as a printed publication because it had been catalogued and shelved in a university library, making it accessible to interested persons exercising reasonable diligence. 781 F.2d 897, 898-99 (Fed. Cir. 1986).¹ Likewise, in In re Hall, the court treated a single thesis catalogued and shelved in a university library as a printed publication because it was publicly accessible.²

Whether a disclosure is “public” for purposes of § 102 also turns on whether it was made subject to meaningful confidentiality restrictions. In Delano Farms, the Federal Circuit held that distributing plant material did not amount to an invalidating public use because the recipients, although not bound by an express confidentiality agreement, were subject to an expectation of secrecy under the circumstances.³ By contrast, unrestricted use of an invention by another person, without meaningful confidentiality restrictions, may support a public-use bar.⁴

The same framework has clear implications for AI tools. If an inventor or attorney discloses unpublished invention details to an AI system that is effectively public, searchable, retained for broader use, reviewed outside a confidential setting or otherwise not subject to meaningful confidentiality restrictions, that disclosure may qualify as public under 35 U.S.C. § 102(a)(1)—creating significant novelty risk. Depending on the circumstances, the issue may arise as a printed-publication bar, a public-use bar or an argument that the invention was otherwise made available to the public.

That said, disclosure to an AI system does not automatically trigger a statutory bar. A confidential disclosure in a genuinely closed AI environment is materially different from entering the same information into a public-facing consumer tool. The critical questions are practical ones: who can access the input, whether the provider may retain or reuse it, whether the applicable terms impose meaningful and enforceable confidentiality obligations and whether the information could later become retrievable by others.

AI Use Also Raises Inventorship and § 112 Risks

Separate from the § 102 issues discussed above, AI-assisted patent drafting also raises inventorship questions. Inventorship still turns on human conception, meaning the formation in the mind of the inventor of a definite and permanent idea of the complete and operative invention.⁵ The Federal Circuit has likewise made clear that only natural persons may be inventors under the Patent Act.⁶ Using AI to help articulate, organize or refine an invention does not eliminate the requirement of human conception. But if the inventive concept exists only because the AI system supplied it, the inventorship analysis becomes considerably more difficult.

AI can also create a subtler prosecution problem by expanding an invention beyond what the inventors actually conceived. In practice, AI can turn a narrower technical contribution into broader and more polished language that makes the invention seem more complete than it really was, creating written-description and enablement risks under 35 U.S.C. § 112. The Federal Circuit has long held that the specification must demonstrate possession of the claimed invention.⁷ And the Supreme Court recently reaffirmed that enablement requires the specification to teach the full scope of the claimed invention without undue experimentation.⁸

In practical terms, AI-assisted drafting can make an invention appear more fully developed than it really was at the time of filing. The more heavily AI is used in drafting the novel portions of an application, the more important it becomes for inventors and counsel to confirm that the application accurately reflects what the inventors actually conceived and possessed as of the filing date.

Litigation Risk Starts During Prosecution

Even if these issues do not arise in examination, they may surface in litigation. An accused infringer will likely ask how the application was drafted, what information was entered into external systems, what the inventors truly conceived before filing and whether the drafting process introduced unsupported terminology, overbroad concepts or undisclosed prior-art references. If discovery uncovers problems in any of those areas, the accused infringer will almost certainly use them to challenge validity, enforceability or both.

There is also a newer risk worth watching. If a generative-AI system draws on prior-art references or technical materials that the drafting team did not identify, future litigants may try to cast that omission as part of an inequitable-conduct theory or, at a minimum, as a credibility problem in discovery. That argument will not fit every case, but it is a risk that should be considered both when drafting the application and during, or when contemplating, litigation. The Federal Circuit has made clear that inequitable conduct remains a serious allegation where material information is withheld with specific intent to deceive.⁹ AI-assisted drafting that functions as a black box can make any later inquiry into what information was used, what was known and what should have been disclosed far more difficult.

Practical Steps to Reduce AI-Related Patent Risk

AI-assisted patent drafting should be managed as a controlled workflow, especially when unpublished invention details are involved. Patent applicants and prosecution counsel should consider the following steps:

• Use closed or contractually protected environments. Do not enter unpublished invention details into public-facing AI tools without first assessing confidentiality.
• Keep humans responsible for conception and claim scope. Inventors and counsel should remain directly involved in identifying the inventive concept, shaping the claim strategy and drafting the portions of the application that define the invention.
• Document the workflow. Maintain a clear record of who contributed what, when the inventors conceived the claimed subject matter, what information was entered into any AI system and what substantive revisions were made by humans.
• Pressure-test support and scope. If AI helped draft claims or specification language, confirm that the inventors actually conceived and possessed the claimed subject matter. Counsel should also confirm that any added context or disclosure introduced during AI-assisted drafting was likewise conceived and possessed by the inventor and is adequately supported in the application.
• Identify the sources the system used. Where possible, require the tool or workflow to identify prior-art references, source materials, or external technical content used to generate drafting language. Any material information identified through that process should then be evaluated for disclosure to the USPTO, including, where appropriate, through an Information Disclosure Statement.

Bottom Line for AI-Assisted Patent Drafting

The principal risks of using generative-AI tools in patent drafting do not arise from any new, AI-specific patent rule. They arise from the application of familiar patent-law doctrines to a new technology. Issues such as novelty, public availability, inventorship and specification support may become more difficult to satisfy, or to defend later, if invention details are entered into AI systems without adequate confidentiality, oversight and documentation.

For businesses and patent counsel, the practical takeaway is straightforward: use AI cautiously, keep sensitive invention details out of public-facing tools and ensure that human inventors and lawyers remain firmly in control of conception, claim scope and drafting.

For more information on patent risks and best practices for using generative AI in invention and prosecution workflows, contact the alert authors or your preferred Polsinelli attorney.