The Federal Landscape: USPTO and the "Human-in-the-Loop"
At the federal level, the United States Patent and Trademark Office (“USPTO”) has taken a firm stance: AI is a tool, not a lawyer.
The 2025 Revised Guidance
Following updated guidance released in late 2025, the USPTO reaffirmed that while AI can assist in the creation of a mark or the drafting of an application, it cannot be an applicant or inventor. Only natural persons can be inventors. AI systems, including generative AI and other computational models, are instruments used by human inventors, analogous to laboratory equipment, computer software, research databases, or any other tool that assists in the inventive process. As the case law establishes, inventors may “use the services, ideas, and aid of others” without those sources becoming co-inventors. (Shatterproof Glass Corp. v. Libby-Owens Ford Co., 758 F.2d 613, 624 (Fed. Cir. 1985) (quoting Hobbs v. United States Atomic Energy Commission, 451 F.2d 849, 864 (5th Cir. 1971)); see also Hess v. Advanced CardiovascularSys., 106 F.3d 976, 981 (Fed. Cir. 1997) (quoting O'Reilly v. Morse, 56 U.S. 62, 111 (1853) (“it can make no difference … whether [the inventor] derives his information from books, or from conversation with men skilled in the science.” … “the fact that Morse sought and obtained the necessary information and counsel from the best sources, and acted upon it, neither impairs his rights as an inventor, nor detracts from his merits.”). The same principle applies to AI systems: they may provide services and generate ideas, but they remain tools used by the human inventor who conceived the claimed invention. (Patent and Trademark Office Docket No. PTO-P-2025-0014).
The conception of a mark—the formation in the mind of the inventor of a definite and permanent idea of the complete and operative invention as it is hereafter to be applied in practice (Burroughs Wellcome Co. v. Barr Labs., Inc., 40 F.3d 1223, 1228 (Fed. Cir. 1994) (citing Hybritech Inc. v. Monoclonal Antibodies, Inc., 802 F.2d 1367))—remains a purely human endeavor. This means that conception is complete when “the inventor has a specific, settled idea, a particular solution to the problem at hand, not just a general goal or research plan.” Id. Analysis of conception turns on the ability of an inventor to describe an invention with particularity. (Burroughs Wellcome Co., 40 F.3d at 1228 (citing Sewall, 21 F.3d at 415)). Absent such a description, an inventor cannot objectively prove possession of a complete mental picture of the invention at a later time. Id.
The Duty of Candor and Reasonable Inquiry
Perhaps the most significant change for practitioners is the emphasis on the Duty of Candor. Under federal rules, any filing submitted to the USPTO carries an implicit certification that the contents are accurate to the best of the filer's knowledge. So, relying solely on AI without human verification is considered a failure of "reasonable inquiry," which can lead to sanctions or the striking of filings. See 37 CFR 4.10. Therefore, if an AI cites fictitious legal precedent, misrepresents a fact, or incorrectly completes a portion of the application, the human signing the document is held responsible.
The State Level: Administrative Efficiency vs. Substantive Rights
While the Lanham Act (15 U.S.C. § 1051 et seq.) provides nationwide trademark protection, state-level registrations (typically via a Secretary of State) offer a different regulatory landscape:
Administrative Filings: State systems are often more administrative than substantive. Uses of AI have significantly streamlined the initial filing process, making "knock-out" searches across all 50 states nearly instantaneous.
Common Law Risks: AI-generated logos and names carry a hidden danger at the state level: unintentional mimicking. Because generative AI is trained on existing data, it may produce a logo that is confusingly similar to a local business that relies on common law rights rather than federal registration.
Pros and Cons of Using AI in Trademarking
AI certainly comes with its multitude of pros; however, it is not without its cons. One advantage to using AI in respect to clearance searches is that AI can scan millions of records in any number of seconds, as well as identify phonetic and visual similarities humans might miss. AI can also help in drafting a trademark application by providing an accurate description of a logo or product. However, a related disadvantage is that AI can miss nuanced conflicts (e.g., conceptual similarities) and may not account for common law marks not in databases. Another advantage is AI can dramatically reduce the hours spent on initial drafting and data collection. However, the costs and time saved using AI could be lost if the applicant receives an Office Action caused by unreviewed AI drafting errors. AI can also assist with multijurisdictional searches by easily comparing a mark against global registries (EUIPO, WIPO, etc.) simultaneously. But AI lacks understanding of local cultural nuances or bad faith filing patterns specific to certain regions, such as the Middle East where it is recommended to secure both the English and Arabic script versions of a trademark. What can happen is a foreign company enters the market with their standard Latin-script logo but neglects to register the phonetic Arabic equivalent. Bad faith actors take advantage of this gap by registering the precise Arabic script spelling of the brand, effectively blocking the true owner from using localized signage, menus, or packaging. AI can also assist with data organization by automating the categorization of goods and services. But AI can struggle with gray areas where a product might fit into multiple classes.
Where AI Fails: The "Likelihood of Confusion" Gap
The core of trademark law is the DuPont factors (In re E. I. du Pont de Nemours & Co.), a subjective analysis of whether two marks are likely to cause consumer confusion. This is where AI currently falls short. In the market context, AI can identify two names that sound alike, but it cannot easily assess the fame of a mark or the specific purchasing habits of a niche consumer base. AI may not provide accurate results when used to assess whether a proposed mark is similar to existing marks. When it comes to strategic advice, AI provides data, but it cannot provide counsel. It cannot advise a client on whether to "fight or fold" when faced with a Cease-and-Desist letter based on the prevailing judicial climate of a specific circuit. Because of these reasons, human review remains necessary when traversing the trademarking process.
Conclusion
AI has transformed trademarking from a game of "searching for a needle in a haystack" to a game of "filtering the results." While the speed and breadth of AI tools are unmatched, the legal weight of a trademark remains anchored in human judgment. For now, the most effective strategy is a hybrid approach: AI for the heavy lifting, and human expertise for the strategic finish. If your company needs assistance with a trademark issue, I invite you to schedule a consultation with Wandzel Law™️