The Death of the 18-Month Publication Rule

Abstract

35 U.S.C. § 122(b) requires that most patent applications be published 18 months after their earliest effective filing date. Enacted in 1999 as part of the American Inventors Protection Act, the rule harmonized U.S. practice with the international standard established under the Patent Cooperation Treaty and the European Patent Convention, and it eliminated the “submarine patent” strategy that had allowed applications to stay secret for decades before surfacing with broad claims.

The rule was designed for a world where examination took three to four years. In that world, 18 months was a reasonable mid-point: early enough to give competitors visibility into pending claims, late enough to give applicants time to assess commercial value before public exposure.

AI to AI prosecution breaks every assumption the rule was built on. When examination completes in days rather than years, the 18-month window is not a useful mid-point. It is an arbitrary delay interposed between a filed application and an examination system that could have processed it weeks earlier.

This article examines why the 18-month rule exists, what the USPTO backlog actually consists of, how an API-first examination architecture would compress the timeline, and what publication framework replaces 18-month publication in a world of near-instantaneous examination.

I. What the 18-Month Rule Does

Under 35 U.S.C. § 122(b), the USPTO publishes patent applications promptly after the expiration of 18 months from the earliest filing date for which a benefit is sought. Publication occurs regardless of whether examination has begun, is in progress, or has concluded.

The published application becomes prior art as of its filing date under 35 U.S.C. § 102(d). This is critical: the prior art date is not the publication date. It is the filing date. So a published application filed in January 2024 and published in July 2025 is prior art as of January 2024 against everything filed after that date.

Applicants may request non-publication under 35 U.S.C. § 122(b)(2)(B) if they certify that the invention has not been and will not be the subject of an application filed in another country that requires publication at 18 months. The non-publication request must be made at filing and can be rescinded if the applicant later decides to file foreign. Approximately 10-15% of applicants make this election.

II. Why It Was Created

Before 1999, the USPTO did not publish applications. Applications remained secret until a patent issued or the application was abandoned. This created the submarine patent problem: an applicant could keep an application pending for years or decades through continuation practice, then surface with broad claims after an industry had unknowingly built on the technology.

Jerome Lemelson famously used this strategy for machine vision and bar code patents, collecting over $1.5 billion in royalties from companies that had no notice of the pending claims when they built their products. The AIPA amended section 122 to eliminate this strategy.

The 18-month window also served a harmonization function. The PCT required publication at 18 months. The EPO published at 18 months. U.S. applicants filing internationally were already subject to 18-month publication through the international route. Making domestic applications subject to the same timeline eliminated the anomaly of different treatment for domestic-only filers.

Both rationales were sound in 1999. The submarine patent problem was real. Harmonization with international practice was a legitimate policy goal. The 18-month window, measured against a 3-4 year examination cycle, was a reasonable compromise between applicant secrecy interests and public notice.

III. The AI Prosecution Problem

The current USPTO reports average first action pendency of 22 to 26 months. Total pendency from filing to grant runs 24 to 36 months. Against that timeline, 18-month publication makes sense: the application has been pending for a year and a half and examination has not yet begun. Publication creates a public record of the pending claims before the applicant reaches the examination stage.

In AI to AI prosecution, first action pendency drops to 24 days. The examination is structurally complete before 18 months has elapsed by 17+ months. There are two scenarios:

Scenario A: Application allowed on first action. The patent grants in approximately 30-60 days (assuming normal formalities). Publication has not yet occurred. Under current law, the patent issues, the application is published simultaneously with grant, and the prior art date is the filing date. This is actually the correct outcome and no reform is needed in this scenario.

Scenario B: Application receives office action; prosecution continues. After the first office action at 24 days, the applicant responds. Additional rounds of prosecution follow. The application may still be in prosecution at 18 months, at which point it is published. In this scenario, the rule functions as intended: publication occurs while the application is still pending.

But Scenario B is where the problem becomes interesting. Under hypothetical AI prosecution, an application receiving an office action at 24 days and a response at 90 days would receive a second office action at approximately 120 days. The prosecution cycle even with multiple rounds completes well before 18 months. The application either grants or goes abandoned before publication occurs.

This creates a situation where the 18-month publication rule publishes applications that have already been disposed of — either by grant or abandonment. Publication of a granted patent is already handled by the patent publication system. Publication of an abandoned application is arguably counterproductive: it creates a public record of claims that were rejected, potentially misleading competitors about what prior art the applicant was pursuing protection for.

IV. The Administrative Delay Problem

Before examining why the 813,000-application backlog exists, it is necessary to understand what actually constitutes the first USPTO communication on a new application.

The first action a new application receives is frequently not a substantive rejection on the merits. It is one of the following:

• Notice of Missing Parts (37 CFR 1.53): Required fee not received, oath or declaration not submitted, specification incomplete, drawings required but not filed.
• Notice of Omitted Items: Pages of specification or drawings missing from the filed application.
• Notice of Informal Application: Claims are not in proper dependent form, specification does not conclude with claims, application not in English.
• Filing receipt with fee deficiency: Underpayment of filing fees or search fees requiring surcharge.
• Restriction requirement: Application claims more than one independent and distinct invention; applicant must elect one for immediate prosecution.

Every one of these is deterministic. Zero legal judgment is required. Every one of these is a checklist operation that can be expressed as a structured API call.

Under current practice, these notices are generated by a human examiner or a USPTO processing specialist, entered into the PALM system, mailed or transmitted via EFS, and the application sits in a queue waiting for the response before substantive examination begins. Elapsed time from filing to resolution of a Notice of Missing Parts: often 3-6 months.

An API-first USPTO would resolve these at the moment of filing. The filing API validates the submission against every formal requirement in 37 CFR Part 1 before acceptance. Missing parts are flagged immediately in the same session. The applicant corrects and resubmits. Resolution time: minutes, not months.

V. Anatomy of the Backlog

The USPTO currently has 813,000 unexamined patent applications in its backlog (FY2025). Total pending applications exceed 1.19 million. Understanding why that backlog exists is necessary before proposing structural remedies.

The USPTO employs approximately 13,000 patent examiners. Each examiner is expected to produce a defined number of “disposals” per year — either allowances, final rejections, or abandonments — measured in production units (PUs) that vary by technology area. A disposal in a complex biotechnology case takes more examiner time than a disposal in a simple mechanical case.

The filing rate has increased materially over the past decade. Utility patent application filings now exceed 650,000 per year. The USPTO’s hiring capacity is constrained by its fee-funded budget (the agency operates on fees set by Congress and collected from applicants) and by the lead time required to train new examiners to journeyman competency, which takes 2-3 years of structured training in most technology areas.

The backlog is therefore structural: filing volume has grown faster than examiner capacity, and examiner capacity grows slowly because training takes years and retention of experienced examiners is a persistent challenge (private practice pays substantially more than GS-scale salaries).

The examination process itself is inherently sequential. A single examiner owns each case from initial examination through disposal. There is no parallel processing. If the examiner is on leave, in training, or managing 400 other cases, the case waits.

Within that sequential structure, the examination workflow involves:

• Initial review of specification, claims, and drawings
• Classification of claims into appropriate patent classifications
• Prior art search across USPTO databases, NPL, and foreign patents
• Analysis of search results against claim elements
• Drafting the office action with specific statutory bases and prior art citations
• Supervisory review before office action mails

Each of these steps is human, serial, and unbounded in calendar time. The case sits in the examiner’s docket alongside hundreds of other cases at various stages. The 22-26 month first action pendency is not the time it takes an examiner to examine a case. It is the time a case waits before an examiner begins examining it, plus the time the examiner spends on the case.

VI. An API-First USPTO

The USPTO currently operates Patent Center, a web-based filing system that replaced EFS-Web. Applications are submitted as XML or PDF packages. The system validates certain formal requirements but routes substantive processing to human examiners.

An API-first USPTO architecture would look fundamentally different. The entry point would be a structured API that accepts claim data, specification text, and drawing files in machine-readable format. At the moment of submission, the API executes:

• Formal compliance check: All required elements present, fees correct, oath executed, dependent claims properly formed. Returns pass/fail with specific deficiency codes. Resolution time: milliseconds.
• Classification: AI-assisted CPC classification of claims. Routes application to correct examination track. Resolution time: seconds.
• Prior art search: Vector search across 18 million U.S. patents, PCT documents, EPO filings, and NPL databases. Returns ranked reference list with claim-element mapping. Resolution time: minutes.
• Statutory analysis: AI applies 35 U.S.C. 101, 102, 103, and 112 frameworks to claims against retrieved prior art. Generates structured office action with specific rejection bases. Resolution time: minutes to hours for complex cases.
• Human review gate: For AI-generated office actions that meet defined confidence thresholds, a supervisory examiner reviews and approves. For complex cases below threshold, full human examination. The human is in the loop as reviewer, not first-pass examiner.

The applicant side of this architecture already exists in commercial form. AI patent prosecution tools are doing real-time prior art mapping, statutory analysis, and response strategy generation against real USPTO office actions today. The infrastructure for structured patent data processing is proven. The gap is on the examination side.

The USPTO has existing programs that point in this direction. The Patents End-to-End (PE2E) initiative and various AI pilot programs within the Office of Patent Legal Administration signal institutional awareness of the opportunity. What has not yet occurred is a structural redesign of the examination workflow from human-first to AI-first with human oversight.

VII. Three Alternative Publication Models

If AI prosecution renders the 18-month publication rule obsolete, three alternative models merit consideration. Each serves the two original policy goals eliminating submarine patents and providing competitor notice but calibrated to an AI examination timeline.

Model A: Publication at Filing (Immediate)

The application is published immediately upon acceptance of the filing. The prior art date remains the filing date. Competitors have notice from day one. No 18-month delay. No submarine patent risk.

The objection to this model is that it forces applicants to reveal their inventions before they know whether they will receive patent protection. If the application is later rejected and abandoned, the applicant has disclosed the invention without receiving anything in return, potentially precluding trade secret protection.

This objection weakens substantially under AI examination. If examination completes in days rather than months, the applicant knows very quickly whether protection is likely. A provisional filing strategy combined with immediate publication of non-provisional applications resolves most of the commercial risk.

Model B: Publication at First Examination Event

The application is published when the first substantive office action or notice of allowance is issued. Under AI examination, this occurs within days to weeks of filing. Publication is tied to an examination event rather than a fixed calendar date.

This model preserves the competitor-notice function. Once the USPTO has assessed the claims, the public knows what protection is being sought. The applicant has some time between filing and publication (days to weeks) to assess commercial strategy.

The prior art date question becomes more complex: should the prior art date be the filing date or the publication date? Current law uses filing date. Under this model, maintaining filing date as prior art date is consistent with existing doctrine and avoids creating gamesmanship around publication timing.

Model C: Publication at Grant or Abandonment

The application is not published until it either issues as a patent or is abandoned. Under AI prosecution, this means publication occurs within weeks to a few months of filing in most cases. The public record reflects only final dispositions, not in-progress prosecution.

This model most closely mirrors how the patent system functioned before 1999, but with the critical difference that submarine patent strategy is foreclosed by the speed of examination. An applicant cannot keep claims secret for decades if examination completes in weeks. The incentive for submarine continuation strategy — keeping claims pending while an industry adopts the technology — disappears when prosecution cycles are measured in months rather than decades.

VIII. Prior Art Date Implications

The prior art implications of publication reform deserve separate treatment because they affect the entire patent system, not just pendency statistics.

Under current law, a published application is prior art as of its filing date under 35 U.S.C. § 102(d). This means that a competitor researching the prior art landscape must search both issued patents and pending published applications. The 18-month publication date tells competitors when to expect a given application to enter the prior art record as a published document, even though its effective prior art date predates publication.

Under AI prosecution with publication at or near examination, the published record becomes more complete more quickly. A competitor searching today’s landscape would find a more current picture of pending applications because the publication delay collapses from 18 months to days or weeks. This benefits freedom-to-operate analysis: the gap between filing date and publication date, during which applications are silent prior art, shrinks dramatically.

The counterargument is interference and derivation concerns. If an application is published before examination is complete, and examination reveals that the claims conflict with another pending application, the derivation proceeding framework under 35 U.S.C. § 135 must accommodate faster-moving applications. This is a tractable procedural reform, not a structural impediment.

The secret prior art problem, i.e., applications that are prior art from their filing date but not yet public, persists under any publication model. Under AI prosecution, the duration of the secret prior art window compresses from 18 months to the time between filing and first examination event (days to weeks). This is a material improvement in the predictability of freedom-to-operate analysis.

IX. Conclusion

The 18-month publication rule is not a constitutional principle. It is a policy compromise calibrated to a specific examination timeline that is about to change fundamentally.

The rule solved two real problems in 1999: submarine patents and international harmonization misalignment. Both problems remain relevant. But both can be solved more precisely by a publication framework tied to examination events rather than a fixed calendar date.

The examination timeline is changing because the applicant side of prosecution has already shifted to AI. The infrastructure for AI-assisted patent analysis — prior art search, statutory analysis, structured response generation — exists and is deployed commercially today. The examination side of the USPTO is the remaining holdout.

When the examination side adopts the same infrastructure, the 18-month publication rule will become a rule that publishes applications that have already been disposed of, either by grant or abandonment. At that point, the rule is not protecting competitors. It is creating administrative overhead for a scenario that no longer exists.

The legislative reform should precede the examination transformation rather than follow it. Patent law scholars, PPAC, and the USPTO Director’s office should be designing the replacement framework now, before the examination transformation makes the current publication rules incoherent.

The 18-month publication rule was a good solution to the problems of 1999. It is not a good solution to the problems of 2030.