USPTO's AI Trademark Toolkit Reshapes Pre-Examination: What Attorneys Must Understand Now

The Tools, Plainly Stated

USPTO, Alexandria, Virginia. March 19, 2026.

The agency announced Class ACT, the Trademark Classification Agentic Codification Tool: an AI agent that automatically assigns international Nice Classification classes, design search codes, and pseudo marks to incoming trademark applications. That trio of tasks had, in recent years, been taking up to five months to complete manually, a delay driven largely by the surge in filing volumes that followed the post-pandemic brand registration boom. Class ACT collapses that window to minutes, sometimes seconds.

In early April 2026, a second tool arrived: an AI-powered image search capability within the USPTO's trademark search system. Users with a MyUSPTO account can upload a design and receive a visual comparison against the entire corpus of federally registered marks, functioning comparably to reverse image search tools on commercial platforms. The agency described it as "a dream come true" for the trademark side. It also released a mark description and colour claim generator, live from April 23, which assists applicants in drafting the precise descriptive language the Office requires for design marks, an area that has historically generated unnecessary office actions through poor drafting alone.

More is coming. USPTO examining attorneys are expected to receive access to an internal generative AI tool this month, July 2026, once the agency's guardrails, rules, and policies clear their final review. The USPTO has also issued two Requests for Information: one seeking vendor solutions to combat trademark fraud, which the agency has publicly acknowledged is a "significant problem", and another seeking vendors to assist with examination and quality review itself.

That is a lot of moving parts in four months.

What Was Actually Decided

None of these tools replace the examining attorney. That point bears emphasis because the headlines have not always made it clear. Human USPTO staff still review Class ACT's output before it becomes operative, and applicants retain the ability to accept, reject, or modify AI-generated suggestions before importing them into their applications. Acting Trademark Commissioner Dan Vavonese put the institutional philosophy plainly: the objective is to free examiners to apply "experienced judgment and reason to the substantive issues in examination."

Acting Chief AI and Data Officer Rob Hayes was characteristically blunt: "AI done right: faster results, higher quality, happier stakeholders."

That framing is accurate as far as it goes. Classification is what legal technologists call a bounded task: well-defined inputs, verifiable outputs, a finite universe of correct answers. Nice Classification has 45 classes. Design search codes operate within a structured taxonomy. These are exactly the conditions under which current AI performs reliably. Contrast this with the USPTO's parallel AI rollout on the patent side, where the AI-assisted prior art search tool ASAP! has drawn pointed criticism from practitioners who found its citations, in the words of one TC 3700 attorney, ranging from tangentially relevant to outright misleading. The difference in reception is not random. Prior art search in emerging technology fields requires judgment about analogous art and claim scope that classification simply does not. Class ACT succeeded where ASAP! has struggled because the USPTO chose the right tool for the right task.

The Practical Impact on Attorneys

For trademark counsel, the immediate consequence is a reordering of the clearance workflow, not its elimination.

The image search tool is genuinely useful for design clearance. Before it existed, identifying visually similar marks required translating visual elements into design search codes, a process that depended on the filer's ability to describe what they could see. The new capability lets you search the register the way a client actually perceives a mark: by looking at it. That is not a marginal improvement. The USPTO itself has flagged it as a beta feature and recommends treating it as a supplemental strategy rather than a replacement for traditional clearance searches. Practitioners who rely on it as a standalone clearance tool are building on sand.

The risks from Class ACT are subtler and more consequential. If the AI assigns incorrect international classes, or generates an inaccurate pseudo mark for an unconventional spelling, the downstream effects compound. An application evaluated against the wrong universe of existing marks may miss real conflicts or trigger spurious ones. The likelihood-of-confusion analysis that follows is only as reliable as the classification that precedes it. Human reviewers at the USPTO catch some of these errors. Not all of them.

Professional responsibility follows the attorney, not the algorithm. A practitioner who accepts AI-generated classifications without independent verification, and whose client later faces a cancellation proceeding rooted in a missed conflict, cannot point to Class ACT as a defence. The tool is the Office's tool. The advice is still yours.

The attorney's value proposition shifts rather than shrinks. Routine classification of straightforward word marks becomes faster and cheaper for clients. The attorney's time moves upstream, into strategic decisions about class selection and the identification of edge cases where AI outputs require scrutiny: marks with abstract design elements, highly stylised scripts, or deliberate phonetic distortion.

A Structural Signal the Market Has Not Fully Priced

AIBD's analysis of Companies House SIC 69.10 filings (law firm incorporations) shows only 61 new entities registered in Q3 2026, a 91.3% decline against the prior period. Read carefully, that figure is not evidence that the legal sector is contracting. New incorporations at this cadence suggest that the boutique IP firm formation wave of recent years has plateaued sharply. The firms that formed during the post-pandemic trademark filing surge are now established. The question going forward is whether workflow automation, led precisely by tools like Class ACT, will concentrate revenue in incumbents who adapt quickly, or whether it will lower barriers enough to produce a second wave of leaner, AI-augmented practices.

History suggests the former. When the USPTO introduced TEAS electronic filing in the late 1990s, the prediction was that it would disintermediate attorneys entirely. Instead, it raised the baseline competence expected of every practitioner and filtered out those who could not adapt. Faster processing surfaces problems sooner. Clients who previously waited months for examination to begin may now discover classification errors, or conflicting marks, within days of filing. That accelerates the advice cycle. It does not eliminate it.

What Comes Next

The generative AI tool for examining attorneys, expected imminently, is the most significant near-term development. Its scope, guardrails, and the extent to which its outputs will be visible to applicants or their counsel remain unpublished. The USPTO has stated that policies will be in place before deployment. Whether those policies address evidentiary questions about how AI-assisted examination decisions are documented and challenged is the question practitioners should be pressing.

The two outstanding Requests for Information on fraud detection and examination assistance will produce vendor contracts. Those contracts will define what external AI systems have access to USPTO examination data, and under what conditions. Watch that procurement process closely.

The USPTO has stated its goal clearly: lead the world on trademark AI. That ambition is admirable. But leading means being first to encounter the failure modes too. The attorneys who understand these tools precisely, including where they break, will be indispensable. Those who treat them as black boxes will eventually produce advice that reflects that ignorance.