Key Takeaway(s): AI tools are reshaping patent preparation and prosecution, but statements made to the USPTO are permanent, and AI-generated inaccuracies don’t come with warning labels. Efficiency gains are real, but so is the need for proper oversight to protect your portfolio.
Recent headlines involving AI‑generated inaccuracies in court filings1 have prompted renewed scrutiny of the use of artificial intelligence across the legal profession. While much of the focus has been on litigation, these developments also raise important considerations for patent prosecution before the USPTO. Patent prosecution is governed by an ongoing duty of candor and good faith, and statements submitted to the Patent Office, including those describing prior art, framing technical distinctions, or summarizing relevant legal principles, are expected to reflect reasonable inquiry and accuracy.2-4 Although courts apply a high bar when evaluating challenges based on prosecution conduct, inaccuracies in the prosecution record can nonetheless introduce uncertainty if later scrutinized, particularly for life science companies whose patent portfolios often underpin core commercial and competitive strategies.
The duty of candor is longstanding and is reflected in both the USPTO’s regulations and the Manual of Patent Examining Procedure (MPEP). As described in MPEP § 2016, allegations of inequitable conduct, when successfully proven, have been treated by courts as having portfolio‑wide consequences, rather than being confined to a particular claim or argument.2 At the same time, modern inequitable conduct jurisprudence imposes demanding requirements, including proof of materiality and intent, and findings of inequitable conduct remain relatively rare. The takeaway for applicants is not that routine prosecution missteps will lead to unenforceability, but rather that statements made during prosecution become part of a permanent record that may later be examined in detail.
Against that backdrop, the expanding use of generative AI tools adds a distinct dimension to this risk landscape.5 AI systems can produce content that appears well‑formed and authoritative, but may be “hallucinated,” incomplete, imprecise, or incorrect without obvious warning to the user, making independent review and verification essential. In the patent prosecution context, this may involve over‑generalized summaries of prior art, mischaracterized teachings of cited references, or technical distinctions that have not been fully reconciled with the underlying disclosures. The issue is not whether AI tools may be used as they are increasingly common and not prohibited, but whether they are deployed with appropriate safeguards, oversight, and accountability.
As adoption accelerates, life science companies should view AI use in patent prosecution as an aspect of broader IP governance and diligence. This includes establishing clear internal policies regarding permissible uses of AI, defining attorney review obligations, and implementing verification protocols tailored to the types of outputs these systems generate. Companies may also benefit from engaging proactively with counsel to assess validation processes, documentation practices, and quality control measures as part of ongoing portfolio oversight. These considerations will play an important role in evaluating the robustness and defensibility of a patent portfolio. In an industry where patents often reflect years of research and substantial investment, maintaining confidence in the integrity of the prosecution process remains a priority that no efficiency gain should compromise.
Editor: Brenden S. Gingrich
1 https://www.nytimes.com/2026/04/21/nyregion/sullivan-cromwell-ai-hallucination.html
2 https://www.uspto.gov/web/offices/pac/mpep/s2016.html; See also, J.P. Stevens & Co. v. Lex Tex Ltd., 747 F.2d 1553, 1561, 223 USPQ 1089, 1093-94 (Fed. Cir. 1984)
3 https://www.law.cornell.edu/cfr/text/37/1.56
Subscribe for News and Updates
