USPTO Launches ‘Class ACT’ AI Agent to Cut Trademark Classification Time from Five Months to Five Minutes

The United States Patent and Trademark Office (USPTO) announced on March 19, 2026, the launch of “Class ACT” — the Trademark Classification Agentic Codification Tool — an artificial intelligence agent designed to automate the pre-processing of trademark applications. According to the USPTO’s official announcement, Director John A. Squires described the tool as capable of completing work that previously took five months “in five minutes or even five seconds.” The launch represents the USPTO’s most significant deployment of agentic AI in trademark examination to date and comes as the Office continues to grapple with structurally increasing application volumes.

Class ACT automates the generation of three distinct pieces of metadata that are essential for trademark searchability and examination. First, international classification: the tool determines which Nice Classification class or classes the goods or services in a given application fall under. Second, design search codes: for applications featuring logos or figurative elements, Class ACT assigns the design search codes that make those marks retrievable in similarity searches. Third, pseudo marks: the tool generates normalized character strings for marks with unconventional spelling or non-standard characters, enabling examiner searches to locate phonetically or visually similar marks that might otherwise be missed. Prior to Class ACT, USPTO staff added all three elements manually — a process that had stretched to roughly five months as a backlog given the surge in trademark filings in recent years.

The analysis published by Sterne, Kessler, Goldstein & Fox notes that while AI-generated classification data is still reviewed by USPTO employees before it becomes part of the official record, the information is made available to examining attorneys and the public almost immediately upon AI generation. This near-instant availability means applicants and their representatives can identify potential classification disputes or search code discrepancies early in the prosecution process — before a formal office action is issued. The human review layer is designed to prevent erroneous AI output from entering the permanent record unchecked, while still delivering the speed benefits to applicants and practitioners.

The practical impact on trademark prosecution is multifaceted. For applicants, faster pre-processing reduces the time between filing and the assignment of a formal filing date with complete bibliographic information — a delay that has frustrated practitioners and applicants alike. For trademark attorneys and agents, the near-immediate availability of AI-assigned classifications and design codes creates a new due-diligence step: verifying whether the tool’s output aligns with the practitioner’s own classification strategy, particularly for goods and services that span multiple classes or involve novel business models. For third parties conducting clearance searches, the improved accuracy and timeliness of design search codes and pseudo marks should increase the reliability of prior art searches, reducing the risk that a confusingly similar mark is missed at the application stage.

The launch fits within a broader context of USPTO modernization under Director Squires. The USPTO’s fiscal year 2026 budget projection estimates trademark application filings will increase by approximately 4.9% over the prior year, continuing a multi-year growth trend that has outpaced staffing capacity in pre-processing workflows. Class ACT is one of several AI initiatives the Office has rolled out: earlier in the patent examination context, the USPTO launched the “ASAP!” pilot program for AI-assisted prior art search, and it has published updated guidance on AI-assisted inventions under 35 U.S.C. § 101. The trademark classification tool’s agentic architecture — in which the AI operates through a series of reasoning steps to assign multiple metadata elements, rather than simply performing a single lookup — reflects the maturation of AI deployment at federal agencies from narrow tools to broader workflow automation.

From an international perspective, the USPTO’s move parallels AI adoption efforts at other major trademark offices. The European Union Intellectual Property Office (EUIPO) has deployed machine learning tools for classification assistance, and the Japan Patent Office (JPO) has incorporated AI into examination support workflows. Class ACT’s full automation of the pre-processing pipeline — combining classification, design coding, and pseudo mark generation in a single agentic workflow — may represent a leading benchmark for how trademark offices globally can address the efficiency demands created by rising filing volumes without proportional increases in examiner headcount.