I. The Threat and Gravity of AI-Generated Preemptive Prior Art

A growing wave of AI-generated "Preemptive Prior Art" is flooding the digital landscape — produced not through physical experimentation or genuine knowledge acquisition, but purely through the probabilistic word sequences and linguistic manipulation of large language models (LLMs). Millions of patent-specification-grade text documents are being generated daily and disseminated across the web without any human intervention or technical verification.

This phenomenon threatens the foundations of the patent system in four distinct ways:

  1. Distortion of the Patent System — The "Library of Babel" Effect: Technically unverified, low-quality AI-generated texts saturate the internet, erecting patent barriers against legitimate human inventors whose genuinely valuable technologies are preempted by vague, machine-generated disclosures.
  2. Deprivation of Inventors' Rights and Erosion of R&D Incentives: The indiscriminate "squatting" of prior knowledge by AI weakens patent incentives for applicants and effortlessly undermines the novelty and inventive step of innovations developed at great human and financial cost, thereby impeding industrial progress.
  3. Exponential Rise in Search and Examination Costs: The social burden placed on patent examiners and applicants to sift through and analyze relevant prior art grows at an unmanageable pace, consuming time and public resources at an unprecedented scale.
  4. Legal Abuse by "Public Domain Trolls": Purely mechanical, technically unreliable repositories such as allpriorart.com and Cloem are being cited as legitimate prior art candidates solely on the basis of their formal public accessibility — a trend that threatens to accelerate technological regression rather than progress.

II. Prior Art Eligibility: "Enablement" vs. "Understandability" — A Comparative Jurisdictional Analysis

For prior art to invalidate a patent claim on grounds of novelty or inventive step, the disclosure must satisfy certain standards from the perspective of a person having ordinary skill in the art (PHOSITA). Global jurisdictions have coalesced around two principal paradigms.

The first is the Enablement-centric doctrine, anchored primarily in the United States. The second is the Understandability / Direct Derivability doctrine, associated with the United Kingdom and the European Patent Office (EPO).

1. United States (USPTO / CAFC): Enablement-Centric Approach

Under 35 U.S.C. § 102, the US patent system places the primary emphasis on whether a PHOSITA could physically reproduce (make) the invention from the prior art disclosure.

  • Presumption of Enablement: Any prior art document that has been publicly disclosed or made available as a printed publication is legally presumed to be operable and enabling. The burden of rebutting this presumption — by establishing non-enablement — falls squarely on the applicant.
  • Anticipation by a Single Embodiment (Agilent v. Synthego, 2025): The Federal Circuit confirmed in 2025 that the enablement requirement for § 102 prior art is a categorically different inquiry from the written description requirements of § 112. Prior art need not enable the full scope of the claims; it is sufficient that a single embodiment can be reproduced without undue experimentation. Proof of efficacy or operational utility is not required.
  • Written Rebuttal Without Expert Declaration (In re Morsa): An applicant may challenge the enablement of prior art through a reasoned written argument identifying facial defects in the document itself — without the expense of an expert affidavit — thereby shifting the burden back to the examiner.
  • Status of Non-Enabled Prior Art in Obviousness (§ 103): Even where a reference fails to anticipate a claim due to non-operability, it may still serve as "prior art for all that it teaches" and be cited in combination for purposes of an obviousness rejection.
⚠ Editorial Note — Character of the Agilent v. Synthego (2025) Holding Agilent v. Synthego (2025) did not establish new legal doctrine; it reaffirmed and restated the pre-existing Anticipatory Enablement principle. More precisely, the enablement of a prior art reference is assessed by combining the disclosure of the reference itself with the skilled person's common general knowledge. Prophetic examples within a prior art document do not, by themselves, render the claimed invention invalid, nor does the presence of a single working embodiment automatically defeat all claims.

2. United Kingdom (UK Courts): Understandability-Centric and Two-Stage Approach

UK courts focus less on mechanical reproduction of working examples and more on the integrity of the information conveyed by the prior art document.

  • Synthon Two-Stage Test (UKHL 2005): The House of Lords bifurcated the novelty inquiry into two distinct stages:
    1. Disclosure: The prior art must, when interpreted through the eyes of a PHOSITA, clearly teach a subject matter that would necessarily infringe the later claim if performed — the so-called "flag-planting" principle. The operative question at this stage is whether the skilled person would understand what is being disclosed.
    2. Enablement: Having understood the disclosed subject matter, the skilled person must be able to actually implement the technology using common general knowledge and only routine experimentation. UK practice demands a higher standard of disclosure — "True & Complete Teaching" — than its US counterpart.
⚠ Editorial Note — The Dual-Stage Structure of the Synthon Doctrine The UK Synthon doctrine establishes a binary analytical framework — strictly separating "Disclosure" from "Enablement" — that cannot be reduced to a single axis of "understandability." Both limbs must be independently satisfied: the prior art must inevitably disclose the claimed subject matter when performed (the disclosure stage), and a skilled person must actually be able to implement that disclosed subject matter using common general knowledge (the enablement stage). Characterizing the doctrine as purely "understandability-centric" risks conflating these two distinct inquiries.

3. European Patent Office (EPO): Direct and Unambiguous Derivability — the Gold Standard

The EPO refuses to deny novelty on the basis of mere technical inference or probability alone. The decisive question is whether the prior art communicates its disclosure clearly and without any residual doubt.

  • The Gold Standard: Novelty is destroyed only where the claimed invention is "directly and unambiguously derivable" from the prior art.
  • Implicit Disclosure and Inevitable Result: For an undisclosed feature to be treated as implicitly disclosed, it must constitute an inevitable and unavoidable consequence of carrying out what is explicitly taught — something that is true beyond any reasonable doubt. Any possibility of an alternative outcome, or a need to supplement the disclosure, shifts the analysis to the inventive step stage.
⚠ Editorial Note — Original Scope of the EPO Gold Standard The EPO "Gold Standard" — the test of "direct and unambiguous derivability" — was originally developed and applied in the context of added matter assessment, i.e., determining whether an amendment extends beyond the content of the application as filed (Art. 123(2) EPC). Applying this standard wholesale to novelty analysis, or extending it as a general principle for excluding AI-generated prior art references, represents a doctrinal expansion beyond the test's original legal context and should be treated as a de lege ferenda argument rather than settled EPO practice.

4. Republic of Korea (KIPO) & Japan (JPO): A Hybrid of Understandability and Enablement

Asian patent offices take the position that a document which is formally accessible to the public but so deficient in disclosure that a PHOSITA cannot reproduce the technology is disqualified from serving as prior art altogether.

  • Exclusion of Incompletely Disclosed References as "Cited Inventions": Under the examination guidelines of both Korea and Japan, a prior art document that fails to disclose the technology completely enough for a PHOSITA to easily work (manufacture and use) the invention may be excluded from serving as the basis for novelty or inventive step analysis — though the application of this principle in practice has not been fully consistent.
⚠ Editorial Note — Practical Consistency in Korean and Japanese Practice While incomplete disclosure in a prior art document may affect novelty and inventive step analysis in Korean and Japanese practice, this cannot be reduced to a single bright-line rule whereby any insufficiently disclosed reference is categorically disqualified as prior art. The precise effect and applicable threshold vary on a case-by-case basis, and application of this principle in examination and tribunal proceedings has not been fully uniform. This caveat is also reflected in the body text above and should be borne in mind when reading the comparative table.

Comparative Summary: Prior Art Eligibility Standards by Jurisdiction

Criterion USA (USPTO / CAFC) UK (Courts) EPO Korea (KIPO) / Japan (JPO)
Central Doctrinal Axis Enablement-centric Understandability-centric Understandability + Direct Derivability Hybrid: Understandability fused with Enablement
Core Legal Test Can a PHOSITA reproduce (make) the invention without undue experimentation? Does the reference clearly teach something that would necessarily infringe the later claim? Is the invention directly and unambiguously derivable from the prior art (Gold Standard)? Can a PHOSITA specifically identify and work (manufacture and use) the claimed invention using the document and common general knowledge?
Minimum Enablement Threshold Enabling a single embodiment is sufficient — full claim scope need not be enabled True & Complete Teaching required Disclosure must be reproducible to the skilled person's satisfaction The skilled person must be able to easily work the invention without undue experimentation
Presumption of Enablement Prior art is legally presumed operable/enabling; applicant bears the rebuttal burden No presumption; determined by individual document interpretation and fact-finding Strict verification of direct and unambiguous disclosure; no inference-based presumption Documents that are too incomplete for reproduction are denied "cited invention" status outright
Role of Non-Operable Prior Art in Obviousness Cannot anticipate (§ 102) but remains valid as a combination reference for obviousness (§ 103) Non-enabling documents excluded from citation even at the novelty stage Excluded from novelty; may inform obviousness analysis in a relaxed form Documents that cannot be worked due to incomplete disclosure are entirely excluded from both novelty and inventive step analysis

III. Conditions for AI-Generated Documents to Qualify as Prior Art — Research and Analysis

Under prevailing patent examination and litigation practice, an AI-generated document must satisfy four traditional and technical requirements in order to qualify as prior art:

  1. Public Accessibility: The document must have been meaningfully indexed and available to the public such that a person of ordinary skill could locate it through a reasonable, diligent search prior to the filing date.
  2. Strict Identity Within a Single Reference: Only what a single prior art document actually teaches — whether expressly or inherently — may be recognized as disclosed. No element may be attributed to the reference beyond the strict scope of what it inherently or explicitly discloses.
  3. Enablement: When combined with common general knowledge, only information that is disclosed in the document at a level sufficient to allow the actual invention to be practiced without undue experimentation should be credited. Disclosures falling short of that threshold must not be treated as enabling, regardless of apparent textual coverage.
  4. Technical Reliability and Reproducibility (No Hallucination): The document must be free from fatal deficiencies endemic to AI output — including hallucinated data, combinatorial errors, and missing operational parameters — which are prerequisites for any meaningful Anticipatory Enablement.

๐Ÿ’ก Emerging Academic and Legal Proposals: The Human-Centric Filter

As machine-generated texts increasingly erode the patent system, a growing body of scholarship argues that bare physical accessibility is an insufficient basis for conferring prior art status. Three distinct theoretical frameworks have emerged:

  • The Human Attention Criterion (Zhang & Hou, 2025):
    Existing case law such as Jazz Pharmaceuticals does not require proof that anyone actually read a given reference. However, this standard is ill-suited to an environment in which AI-generated content saturates the web. Zhang and Hou propose that for an online publication to qualify as prior art, the challenging party must affirmatively demonstrate — through objective evidence of digital engagement — that "actual human attention" was directed to the document before the filing date.
  • The Conception Requirement (Lucas R. Yordy, 2021):
    AI systems are no more capable of understanding the technical significance or inventive completion of their outputs than a "monkey typing on a typewriter." Accordingly, Yordy argues that for an AI-generated document to carry invalidating effect, additional empirical evidence — such as implementation data — demonstrating that actual human conception and cognitive recognition of the technical content occurred must be submitted during invalidity proceedings.
  • Substantive Nexus to Human Understanding (John Villasenor, 2024):
    The fictional legal construct of PHOSITA is inherently grounded in the realistic knowledge base of human beings. It follows that a computer-generated text — published through probabilistic token selection without any substantive human engagement — should be categorically excluded from the definition of a "printed publication" under patent law, as it lacks the essential nexus to human understanding that animates the prior art doctrine.
⚠ Editorial Note — Human-Centric Proposals vs. Existing US Public Accessibility Law The proposals advanced by Zhang & Hou (2025), Yordy (2021), and Villasenor (2024) are scholarly and legislative proposals for doctrinal reform, not statements of existing law. Under current US public accessibility doctrine — as confirmed by Jazz Pharmaceuticals and related cases — proof that any specific person actually read a given reference is not required. The operative question is whether a PHOSITA could have located and understood the document through reasonable efforts. The conclusion that AI-generated documents cannot qualify as prior art absent actual human attention does not follow from existing case law; it represents a policy argument for how the law should evolve.

IV. Proposed Reform: Four Eligibility Criteria for AI Mass-Generated Prior Art

To defend the intellectual territory of legitimate human inventors against the flood of AI-generated Preemptive Prior Art and to correct the distortions it introduces into the patent system, the following four prior art eligibility requirements are proposed, synthesizing the doctrinal analysis above.

Proposal 1

Require Proof of Meaningful Indexing and Human Cognitive Engagement for Public Accessibility

Mere retrievability via a Google search query should not suffice to establish public accessibility. Building on the "shoebox card" principle from In re Cronyn — which denied prior art status to inadequately indexed documents — the challenging party should bear the strict burden of demonstrating that, prior to the filing date, a PHOSITA actually directed cognitive attention to the document through a meaningfully indexed and practically navigable search path. Furthermore, repositories contaminated by hallucinations and nonsensical outputs — which offer no reasonable expectation that a skilled artisan would rely on them for technical development — should be categorically denied "Prominent Forum" status and excluded as sources of eligible prior art.

Proposal 2

Codify Immediate Collapse of the Presumption of Enablement Upon Identification of Facial Defects

AI-generated databases discharge text without physical verification and should not benefit unilaterally from the presumption of operability. Where an applicant identifies even a single facial defect in an AI-generated document — such as missing preprocessing parameters, numerical contradictions, or physically impossible technical specifications — the In re Morsa doctrine should operate to immediately collapse the presumption of enablement. The burden of establishing operability must then shift back to the examiner or invalidity claimant who introduced the reference.

⚠ Editorial Note — Scope of the In re Morsa Doctrine In re Morsa demonstrates that an applicant may challenge a reference's enablement through reasoned written argument — without a costly expert declaration — but characterizing this as meaning that identification of even a single facial defect immediately and automatically shifts the full burden of proof is an oversimplification. In practice, the examiner considers the reference together with the skilled person's common general knowledge before reassessing enablement, and whether the burden actually shifts depends on the persuasiveness and specificity of the argument, determined on a case-by-case basis.
Proposal 3

Bar Non-Enabled AI-Generated Documents from Serving as Obviousness Evidence

Although MPEP § 2121 permits non-enabling references to be cited "for all that they teach" in an obviousness context, the indiscriminate application of this rule to hallucination-ridden AI mass-text imposes an unreasonably high obviousness bar. Accordingly, any AI-generated document determined to be non-enabled — in the sense that it cannot directly anticipate the claimed invention due to technical errors or hallucinations — should be excluded entirely as a combination reference in an obviousness analysis, on the grounds that it cannot provide a PHOSITA with a reasonable expectation of success in achieving the claimed invention.

⚠ Editorial Note — Conflict with Current US Law on Non-Enabled Prior Art in Obviousness Under current US law — as established in MPEP § 2121 and supporting case law — a prior art reference that is non-enabled may nevertheless be cited "for all that it teaches" in support of an obviousness rejection under § 103. Proposal 3's call for the wholesale exclusion of such references from obviousness analysis conflicts with this settled principle. It is more accurately understood as a legislative or policy reform proposal rather than a restatement of existing law. Agilent v. Synthego (2025) did not alter this position.
Proposal 4

Adopt the EPO Gold Standard Globally to Filter AI-Generated Noise

Given the sheer volume of numerical ranges, arbitrary parameter combinations, and technical noise discharged by AI systems, EPO's demanding Gold Standard should be broadly adopted. Only disclosures from which the claimed result would follow as a clear and inevitable consequence — beyond any reasonable doubt — upon faithful implementation of the prior art should be recognized. Any AI-generated document that requires supplementation by common general knowledge, additional trial and error, or the addition of genuinely valuable technical information to enable reproduction should be wholly disqualified as prior art.

References

The following sources form the doctrinal backbone for the legal analysis of AI prior art non-enablement and public accessibility presented in this article.

PA 1 · Agilent Technologies, Inc. v. Synthego Corp.
Case Law — U.S. Court of Appeals for the Federal Circuit
Agilent Technologies, Inc. v. Synthego Corp., Nos. 2023-2186, 2023-2187 (Fed. Cir. June 11, 2025).
In a CRISPR-Cas9 patent dispute, the Federal Circuit reaffirmed that the enablement standard for § 102 prior art is lower than the § 112 written description requirement, and that enabling a single embodiment without undue experimentation is sufficient to anticipate a claim.
PA 2 · Attention Is All You Need: Prior Art in the Age of AI
Academic Article — SSRN Electronic Journal
Zhang, A., & Hou, S. M. (2025). Attention is all you need: Prior art in the age of AI. SSRN Electronic Journal.
Proposes that prior art status for online publications should require the challenging party to demonstrate, through objective digital engagement evidence, that actual human attention was directed to the document before the filing date — a direct response to AI prior art flooding.
PA 3 · The Library of Babel for Prior Art
Academic Article — Vanderbilt Law Review
Yordy, L. R. (2021). The Library of Babel for prior art: Using artificial intelligence to mass produce prior art in patent law. Vanderbilt Law Review, 74.
Likens AI's probabilistic text generation to a "monkey typing on a typewriter" and argues that invalidity proceedings should require affirmative proof of actual human conception and recognition of the technical subject matter in AI-generated prior art.
PA 4 · Ten Thousand AI Systems Typing on Keyboards
Academic Article — Vanderbilt Journal of Entertainment & Technology Law
Villasenor, J. (2024). Ten thousand AI systems typing on keyboards: Generative AI in patent applications and preemptive prior art. Vanderbilt Journal of Entertainment & Technology Law, 26.
Demonstrates that machine-generated text lacking a substantive nexus to human understanding should be excluded from the definition of a "printed publication" under patent law.
PA 5 · USPTO Request for Comments on the Impact of AI on Prior Art and PHOSITA
Government Administrative Document — Federal Register
United States Patent and Trademark Office. (2024). Request for comments on the impact of the proliferation of artificial intelligence on prior art and PHOSITA. Federal Register, 89.
Official USPTO public solicitation seeking comments on how AI-generated hallucinations and contradictory disclosures affect the assessment of prior art enablement and operability.
PA 6 · USPTO MPEP § 2121 — Prior Art; General Level of Operability
Examination Guideline — Manual of Patent Examining Procedure
United States Patent and Trademark Office. (2024). MPEP § 2121: Prior art; general level of operability required to make a prima facie case.
Establishes that printed publications and other prior art documents are legally presumed to be operable and enabling, placing the burden of proving non-enablement on the patent applicant.
PA 7 · In re Morsa
Case Law — U.S. Court of Appeals for the Federal Circuit
In re Morsa, 713 F.3d 104 (Fed. Cir. 2013).
Held that a non-enablement challenge may be sustained through a reasoned written argument identifying facial defects in the prior art document — such as missing specific conditions or parameters — without requiring a costly expert declaration.
PA 8 · Jazz Pharmaceuticals, Inc. v. Amneal Pharmaceuticals, LLC
Case Law — U.S. Court of Appeals for the Federal Circuit
Jazz Pharmaceuticals, Inc. v. Amneal Pharmaceuticals, LLC, 895 F.3d 1347 (Fed. Cir. 2018).
Established that public accessibility does not require proof that any specific person actually read the reference; it is sufficient that a PHOSITA could have located and understood the document through reasonable efforts.
PA 9 · Blue Calypso, LLC v. Groupon, Inc.
Case Law — U.S. Court of Appeals for the Federal Circuit
Blue Calypso, LLC v. Groupon, Inc., 815 F.3d 1331 (Fed. Cir. 2016).
Provided the analytical framework for determining the public accessibility of online non-patent literature: a document qualifies if it was (1) meaningfully indexed by an internet search engine, or (2) posted in a "prominent forum" accessible to persons of ordinary skill in the relevant art.

Related Video Commentary

[YouTube] The Threat of AI Mass-Generated Prior Art and Strategies for Defense