Navigating The Risks Of AI-Assisted Patent Drafting

The Growing Role of AI in Patent Drafting and Prosecution

Generative artificial intelligence tools are increasingly being integrated in legal workflows, including patent practice. AI tools now exist to assist patent practitioners with tasks such as summarizing invention disclosures, drafting specifications, generating claim language, and preparing responses to office actions. When generative AI tools are incorporated into the patent drafting process, practitioners and inventors should address several potential risks relating to confidentiality, inventorship, accuracy, and downstream litigation implications. As seen in recent court decisions surrounding AI-confidentiality, these considerations will be important in future patent-related litigation.

Confidentiality Considerations

A key issue associated with AI-assisted drafting is the protection of confidentiality. The process of preparing a patent application involves the disclosure of sensitive information, including invention disclosures, technical specifications, and product development documentation. These materials frequently also contain trade secrets, confidential business data, and other information that is not public until the patent application is officially filed.

When such information is entered into generative AI tools, particularly through the use of publicly available or consumer-grade AI platforms, questions arise as to whether the information remains confidential. From a patentability standpoint, maintaining confidentiality before filing is imperative to prevent public disclosure of an invention. Under 35 U.S.C. § 102(a)(1) claimed inventions must not be published or available to the public before the effective filing date. Although the America Invents Act provides a limited one-year grace period for certain disclosures made by the inventor, ¹ a joint inventor, or a person who obtained the subject matter from the inventor, its application to disclosures made through AI platforms remains uncertain in practice. Key jurisdictions, including Europe, also apply absolute novelty rules that can bar patentability if an invention becomes publicly available before a patent application is filed. ²

AI platforms vary in how they handle the confidentiality of user inputs. Most enterprise AI platforms offer contractual confidentiality protections which limit or prohibit the use of prompts and outputs for model training. However, non-enterprise versions of publicly available AI tools (e.g. Claude, ChatGPT, or Gemini) process inputs on external servers and often maintain terms of service allowing the tool to retain prompts or outputs for purposes such as model training or sharing with third-parties. Entering invention disclosures into such tools could potentially expose confidential information to third-party service providers. The issue of whether such disclosure would constitute public disclosure under § 102(a)(1) is an open question that courts are soon to face.

Recent litigation has already begun to examine the consequences of using these public AI tools in legal contexts. For example, in Warner v. Gilbarco, Inc. ³ in the Eastern District of Michigan, the Court addressed 1) whether AI-generated materials by a pro se plaintiff using ChatGPT in connection with the litigation were discoverable, and 2) whether the AI-generated materials were protected by the work product doctrine. In Warner, the defendants sought broad discovery into the plaintiff's use of AI tools, including prompts, outputs, and related materials, arguing that any protections were waived by disclosure to a third-party platform. The Court rejected that position, holding that the plaintiff's AI-assisted materials constituted protected work product because they were prepared in anticipation of litigation, and further concluding that use of ChatGPT did not effect a waiver as it would not result in disclosure "to an adversary or in a way likely to get in an adversary's hand." ⁴ The Court further noted that "ChatGPT (and other generative AI programs) are tools, not persons, even if they may have administrators somewhere in the background." ⁵

The decision provides an indicator that courts will carefully examine how AI platforms function—including terms of service, data practices, and user activity—when determining if information has effectively been disclosed. As a result, reliance on consumer-grade public AI tools during patent drafting introduces the risk that technical disclosures, if deemed accessible to third parties or insufficiently protected, would be characterized as public disclosures or otherwise undermine confidentiality, depending on the platform and circumstances of use.

While case law is beginning to develop, practitioners, organizations, and inventors should err on the side of caution and consider the potential implications of public disclosure in patent drafting when using generative AI platforms. The following practical considerations will help practitioners and organizations realize the benefits of AI-assisted patent drafting while managing potential risks:

1. Carefully evaluate the confidentiality provisions and data-handling practices of any AI system used in connection with patent drafting, prioritizing enterprise-grade platforms that offer enforceable commitments regarding confidentiality, limitations on model training with customer data, and appropriate data residency protections.
2. Establish clear internal policies and implement input controls specifying when and how generative AI tools may be used with invention disclosures and patent applications, including explicit prohibitions on using consumer-facing AI tools for invention disclosures or specification drafting.
3. Adopt an organizational governance framework that addresses the full lifecycle of AI use, including vendor diligence, contracting, and policy implementation.
4. Maintain written policies and usage logs documenting AI practices, which can be valuable if the confidentiality of an invention is later challenged in litigation or post-grant proceedings.

Hallucinated Content in AI-Generated Applications

Another concern with AI-assisted patent drafting is that generative AI might introduce claims to technical content that was not actually disclosed by the inventor. Modern generative AI tools can produce detailed technical explanations, alternative embodiments, and additional claim language based on relatively limited prompts. While this capability can help practitioners quickly expand a draft specification or explore possible claim strategies, it can also introduce the risk that the model hallucinates and generates inventive material beyond what was disclosed.

By way of example, an AI system prompted to "expand the specification to include alternative embodiments" may generate additional architectures or functional features that were not contemplated by the inventors. If such language is incorporated without verification, the resulting application may describe subject matter that the named inventors did not conceive or enable. If these materials are included in the submission to the USPTO, questions would arise as to whether the inventors themselves actually conceived of the subject matter described within the application. If such material makes its way into the claims, it will create issues with the written description requirement under 35 U.S.C. § 112 which requires that the specification demonstrate that the inventors were in possession of the claimed invention at the time of filing.

As illustrated in Thaler v. Vidal, ⁶ the Federal Circuit has confirmed that the Patent Act requires inventors to be natural persons and that an artificial intelligence system cannot be named as an inventor. Because inventorship in U.S. patent law turns on conception by identified human inventors, the incorporation of AI-generated material complicates the analysis regarding which aspects of the claimed invention were actually conceived by those individuals. The USPTO has similarly emphasized in its Revised Inventorship Guidance for AI-Assisted Inventions that, although AI may be used as a tool, inventorship must remain grounded in significant human contribution, and practitioners must ensure that the claimed subject matter reflects the inventors' own conception. ⁷

For these reasons, verification of any AI-generated content incorporated into a patent application should be required to ensure review by the inventors or by technical personnel who are familiar with the invention and experts on the relevant subject matter. This review step is imperative to confirm that the technical substance remains accurate and attributable to the inventors. Practitioners should be cautious of AI-generated information about the invention that is not found in inventors' disclosures.

Organizations will also benefit from maintaining clear documentation of inventor contributions and the technical sources underlying the content of patent applications, including retention of prompt and output logs in a privileged repository where appropriate. Ongoing training and awareness for inventors and practitioners may also support consistent and appropriate use of AI tools as technologies and associated risks continue to evolve. Used in this manner, AI tools can serve as drafting aids that help expand and refine disclosures while the ultimate technical content of the application remains grounded in the inventors' contributions.

Looking Ahead: Responsible Integration of AI in Patent Drafting

As generative AI tools continue to develop, their role in patent practice will continue to expand. There is no doubt that AI-assisted drafting tools can offer immense benefits to workflows for practitioners and the organizations who allow practitioners to utilize them. However, as practitioners and organizations experiment with AI tools in patent practice, they must implement internal governance frameworks to guide how these tools are used and safeguard their patents against these risks. Thoughtfully integrated AI systems can play a valuable role in patent practice, streamlining the organization of technical disclosures, facilitating the exploration of alternative claim language, and enhancing overall drafting efficiency. Yet, by maintaining conscientious safeguards and ensuring diligent human oversight, organizations can harness these efficiencies without compromising the core responsibilities of inventors and practitioners. Ultimately, when deployed responsibly, AI technologies offer the promise of a more effective and secure patent drafting process that upholds both the integrity and confidentiality of inventive contributions.

This article provides general information on legal topics and is not legal advice.

Footnotes

1. 35 U.S.C. § 102(b)(1).

2. European Patent Convention art. 54 (requiring absolute novelty for patentability for participating member states).

3. Warner v. Gilbarco, Inc., No. 24 Civ. 12333, 2026 WL 373043 (E.D. Mich. Feb. 10, 2026)

4. Id. at *4.

5. Id.

6. 43 F.4th 1207 (Fed. Cir. 2022).

7. Revised Inventorship Guidance for AI-Assisted Inventions, USPTO (Nov. 28, 2025).

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.