The USPTO’s AI Agenda: Examining the Office’s AI Tools and Guidance for Practitioners

The U.S. Patent and Trademark Office (USPTO) is going through a significant digital transformation. With the Office seemingly updating its procedures as rapidly as the latest AI model, it’s important to track what this means for IP practice. AI is transforming the tools governing how the Office now processes what is filed, and the Office’s vacillations on AI inventorship should be top of mind for every practitioner.

The Inventorship Anchor: Thaler and the Squires Rescission

Every current USPTO guidance document on AI inventorship operates within a constraint fixed by federal courts, not by any administration: AI cannot be named as an inventor. In Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022), the Federal Circuit held that inventors must be natural persons under 35 U.S.C. § 100(f). The Supreme Court declined review, and Thaler remains binding precedent.

On February 13, 2024, the USPTO issued Inventorship Guidance for AI-Assisted Inventions (89 Fed. Reg. 10,043), which borrowed the three-factor test from Pannu v. Iolab Corp., 155 F.3d 1344 (Fed. Cir. 1998), a doctrine designed for joint inventorship disputes among humans, and applied it to evaluate a single person’s interaction with an AI tool. Examiners were directed to scrutinize whether prompts constituted a “significant contribution” to each claim, effectively treating AI as an unnamed silent co-inventor.

On November 28, 2025, Director Squires issued Revised Inventorship Guidance for AI-Assisted Inventions (90 Fed. Reg. 54,636), rescinding the 2024 guidance in its entirety. The operative shifts are clear: Pannu applies only when multiple natural persons collaborate; the uniform standard is now conception; and AI is characterized as a tool analogous to laboratory equipment or research databases. The presumption of human inventorship has been restored. The USPTO will not question inventorship at the examination stage unless an AI system is expressly listed as an inventor.

Pannu itself has not been abolished. When multiple natural persons collaborate, the factors continue to govern the joint inventorship analysis. But the more significant risk lies in the courts. The revised guidance binds patent examiners; it does not bind district courts, the Patent Trial and Appeal Board (PTAB), or the Federal Circuit. The outer boundary of what conception requires when AI systems do the heavy analytical lifting remains an open Federal Circuit question.

ASAP! and the Gap Between Ambition and Execution

The inventorship guidance governs what practitioners submit to the USPTO. A separate set of developments governs what happens inside the USPTO when it receives those submissions. Director Squires has made operational AI deployment a stated priority, and the Artificial Intelligence Search Automated Pilot Program (ASAP!) is the first of many similar programs.

Launched in October 2025, ASAP! delivers AI-generated prior art search results to applicants before examination begins. The tool uses an application’s Cooperative Patent Classification (CPC) classification, specification, claims, and abstract to search U.S. patents, pre-grant publications, and foreign patent documents, then sends up to ten ranked documents to the applicant in an Automated Search Results Notice (ASRN). The ASRN is not a 35 U.S.C. § 132 action and requires no response. Its purpose is to inform early decisions on preliminary amendments, deferred examination, or abandonment before the first office action issues.

What is less well known is that this is not the USPTO’s first use of AI in the search process. The Office has operated an AI-based Similarity Search tool inside its internal platform since at least 2022, allowing examiners to identify related patent documents using specification text rather than keyword queries alone.

The participation numbers suggest practitioners are not yet convinced the output quality justifies the participation. As of April 16, 2026, only 169 petitions had been received across all tech centers and only 76 granted, against a program with 3,200 slots. The Office waived the $450 petition fee in March 2026 and extended the petition deadline to June 1, 2026, to boost enrollment.

The IPWatchdog community feedback is direct. One attorney with a TC 3700 grant reported that three of the ten ASRN citations were “at best tangentially relevant” and the remaining seven “missed the mark by miles” and would not support reasonable Section 102 or 103 rejections. A second commenter noted that “USPTO management has been repeatedly told by examiners that their AI search is ineffective and they don’t seem to care.” A third flagged a structural concern: “Not nearly enough information has been made public about the type/methodology/details of precisely what the USPTO is doing with the use of AI internally.”

On the Trademark Side: Class ACT

The Office’s AI deployment is not limited to patent examination. In March 2026, the USPTO launched the Trademark Classification Agentic Codification Tool (Class ACT), an AI assistant that can immediately assign international classes, design search codes, and pseudo marks to unclassified trademark applications, a task that previously took several months.

Acting Trademark Commissioner Dan Vavonese framed the tool’s role clearly: “With the development and implementation of AI tools in our workflow, our employees can focus on applying their experienced judgment and reason to the substantive issues in examination.” Human examining attorneys still review Class ACT output before it becomes operative. Unlike ASAP!, Class Act has seen a successful rollout.

The contrast between Class ACT and ASAP! is instructive. Classification is a bounded, structured task with well-defined inputs and verifiable outputs. Prior art search in emerging technology fields requires judgment about analogous art, claim scope, and technical context that is provided by a human-in-the loop. The difference in early reception reflects that distinction.

Digital Transformation and a Durable Uncertainty

Despite the pace of change at the USPTO (e.g., new AI inventorship standards, ASAP!, Class ACT, and a pro-patent examination posture across multiple guidance documents), the Office’s administrative arc and the judicial arc remain two separate things. Practitioners should recognize both.

The substantive shifts described in the last 21 months are real and operative. Examiners are applying them today. But examination guidance binds examiners, not courts. How the Federal Circuit will evaluate the validity of patents granted under these guidance cycles remains an open question, and opposing counsel in future litigation will not be bound by the same interpretive posture the Office has adopted. The same changes that ease prosecution today may become contested flashpoints in post-grant proceedings or district court litigation tomorrow.

The practical standard for portfolio-level practice has not changed: structure every application to survive the current guidance cycle and the cycle after it. Claim language, specification depth, and inventor documentation should be built for judicial scrutiny, not just examiner scrutiny. The Office may continue to liberalize; the Federal Circuit may not follow.