Generative AI can produce vast quantities of technical language and combinatorial disclosures at very low cost. If such material is published to foreclose future patent rights, it may increase the burden of prior-art searching and patentability analysis. This article uses the terms “mass-generated AI prior art” and “preemptive defensive publication” as descriptive labels for that possibility.
Neither expression is an established, freestanding legal category. Nor is there reliable evidence that millions of patent-specification-grade AI documents are already being published and indexed every day. The issue is therefore best framed as an emerging institutional risk—not as a demonstrated collapse of the patent system.
I. The Core Distinction: Generative Capacity Is Not Legal Prior-Art Status
Projects such as All Prior Art and Cloem have illustrated how software can recombine claim elements, vary technical language, and generate large collections of putative disclosures. Modern generative models can reduce the cost of producing such text even further. But the capacity to generate and upload words does not establish that every output is legally effective prior art.
Authorship by a human or an AI system may be relevant to provenance and reliability, but it should not automatically determine legal status. Human-authored publications can be speculative, erroneous, or non-enabling. AI-assisted documents can be specific, expert-reviewed, and reproducible. A sound inquiry therefore focuses on the objective properties of the asserted reference:
- Was the material publicly accessible before the critical date?
- What technical teaching would a person skilled in the art understand from the document?
- Does one reference disclose every limitation of the claim?
- Could the skilled person practise that disclosure without undue experimentation or burden?
- For inventive-step or obviousness purposes, what does the document reliably teach?
The USPTO’s 2024 request for public comment expressly asked how the proliferation of AI might affect prior art, the knowledge of a person having ordinary skill in the art (PHOSITA), and patentability determinations. That initiative did not declare AI output categorically eligible or ineligible. It recognized that existing doctrine may need careful application in a high-volume environment.
II. A Common Comparative Framework Beyond “Enablement vs. Understandability”
It is misleading to say that the United States is concerned only with enablement while the United Kingdom and the European Patent Office are concerned only with understandability. Although terminology and procedure differ, novelty analysis across these systems generally addresses four related questions:
- Prior-art eligibility: Was the material available to the public before the relevant date?
- Content of the disclosure: What would the skilled person derive from the document as a whole, using the common general knowledge of the time?
- Novelty comparison: Does a single reference disclose every claim limitation expressly or by necessary implication?
- Enablement or reproducibility: Could the skilled person carry out the disclosed teaching without undue experimentation or burden?
The principal differences concern how each system expresses the disclosure inquiry, how demanding the reproducibility analysis is in a particular technical field, and how presumptions and evidential burdens operate.
1. United States: Public Accessibility, Single-Reference Disclosure, and Enablement
An online item qualifies as a “printed publication” under 35 U.S.C. § 102 only if it was sufficiently accessible to the interested public before the critical date. Proof that a particular person actually read the document is not invariably required. The inquiry asks whether interested skilled persons, exercising reasonable diligence, could have located it in light of its distribution, indexing, repository, audience, and surrounding circumstances. Search-engine indexing can be important, but it is not an absolute prerequisite in every case.
To anticipate a claim, one reference must disclose every limitation, expressly or inherently, and must enable at least one embodiment within the claim. The skilled person must be able to carry out the claimed product, process, or use without undue experimentation. The reference need not enable every possible embodiment across the full breadth of a later claim, and proof of ultimate commercial efficacy is not invariably required.
The US “presumption of enablement” should be described carefully. It is not a substantive privilege attached automatically to every publicly available document. During examination, once the examiner makes a proper prima facie case that a reference expressly anticipates or renders obvious all claim elements, the examiner need not independently prove enablement at the outset; the applicant then bears a burden of producing a persuasive non-enablement rebuttal. Where the reference appears non-enabling on its face, however, reasoned argument based on the document itself may suffice without an expert declaration. MPEP § 2121, In re Antor Media, and In re Morsa reflect this procedural structure.
In Agilent Technologies, Inc. v. Synthego Corp. (Fed. Cir. 2025), the Federal Circuit confirmed that a prophetic example or the absence of actual reduction to practice does not automatically defeat prior-art enablement. The reference is read together with the knowledge of the skilled person, and the relevant question is whether an embodiment within the claim could be carried out without undue experimentation.
A reference that is not sufficiently enabling to anticipate may still be considered for what it reliably teaches in an obviousness analysis. Hallucinated, contradictory, or technically impossible portions may have little or no probative value regarding a motivation to combine or a reasonable expectation of success.
2. United Kingdom: Synthon’s Distinct Disclosure and Enablement Requirements
Under the UK approach associated with Synthon, anticipation requires both prior disclosure and enablement. The disclosure inquiry asks whether performing the earlier teaching would necessarily result in subject matter falling within the later claim. The enablement inquiry asks whether the skilled person could actually perform that teaching using common general knowledge.
The doctrine therefore cannot be reduced to an “understandability-centric” test. Nor does Synthon justify a categorical claim that UK law always requires a higher or more “complete” teaching than US law. The systems use different formulations, but both require an enabling disclosure for anticipation.
3. European Patent Office: Direct and Unambiguous Disclosure Plus Reproducibility
At the EPO, claimed subject matter lacks novelty only if it is directly and unambiguously derivable from the prior-art document as a whole, including features necessarily implicit to the skilled person. A feature is not disclosed merely because it is one possible option among many, or because the reader can construct it by retrospectively selecting elements from multiple independent lists.
The phrase “gold standard” is most prominently associated with added-matter analysis, but direct and unambiguous disclosure is also an established novelty criterion. Added matter and novelty nevertheless arise under different provisions and involve different comparisons; they should not be treated as identical legal questions.
EPO novelty doctrine also requires reproducibility. A disclosure is novelty-destroying only if the skilled person, using the common general knowledge at the relevant date, can carry out the technical teaching. The EPO approach therefore includes both the content of the disclosure and whether that content can be practised.
4. Republic of Korea: Identifying a Concrete Cited Invention and Assessing Its Evidential Value
Korean practice asks what technical subject matter the skilled person can objectively identify from the earlier document in light of the common technical knowledge at the filing date. For novelty, the claimed combination must be concretely disclosed in a single cited invention. A document so incomplete that its technical teaching cannot be identified or carried out will ordinarily be a weak or insufficient basis for a novelty rejection.
It is nevertheless too broad to say that every incomplete reference loses all prior-art status or must be excluded from inventive-step analysis. Its relevance depends on the technical teaching it actually conveys, the common knowledge of the skilled person, and any proposed combination with other art. Current KIPO guidelines and fact-specific Patent Court or Supreme Court decisions should be consulted in an actual dispute.
5. Japan: Identifiable Cited Invention and Reliable Internet Disclosure
Japanese examination likewise considers whether a cited invention can be identified from the earlier document by the skilled person using common general knowledge. For internet disclosures, the date, content, accessibility, and reliability of the online record require careful confirmation.
Japanese law should not be collapsed into a single “Asian hybrid” category with Korean law. A reference that does not disclose an enabling or identifiable anticipation may still contain technical information relevant to inventive step, depending on the facts. The JPO Examination Guidelines and Handbook should be applied separately.
Comparative Overview
| Issue | United States | United Kingdom | EPO | Republic of Korea | Japan |
|---|---|---|---|---|---|
| Basic structure | Public accessibility + all claim limitations + enablement | Disclosure + enablement as distinct requirements | Direct and unambiguous disclosure + reproducibility | Concrete identification of cited invention + ability to understand or perform | Identification of cited invention + reliable public disclosure |
| Novelty focus | One reference must enable every claim limitation | Performance of the reference must necessarily fall within the claim and be enabled | All features must be directly and unambiguously derivable and reproducible | The claimed combination must be identifiable in one cited invention | The claimed invention must be identifiable from the cited disclosure |
| Distinctive feature | Prima facie examination framework and rebuttal burden | Synthon separates disclosure from enablement | Express emphasis on direct and unambiguous derivability | Cited invention construed with common technical knowledge | Specific guidance for internet-disclosed technical information |
| Non-enabling reference in inventive-step analysis | May be used for what it reliably teaches | Depends on the actual technical teaching | Depends on disclosed content and technical plausibility | No categorical exclusion; fact-specific assessment | No categorical exclusion; fact-specific assessment |
III. Four Source-Neutral, Evidence-Based Criteria for AI-Generated Documents
Rather than treating AI provenance as a status-based ground for automatic inclusion or exclusion, patent authorities can apply four criteria that are equally relevant to human-authored and machine-generated documents.
Verifiable Public Accessibility
Determine whether the document was meaningfully available to the relevant public before the critical date. Consider indexing, technical classification, repository structure, distribution to the relevant community, access restrictions, reliability of the publication date, and continued availability. Evidence of actual human readership may support accessibility, but it is not a general requirement under current US law.
Integrity and Transparent Provenance
Ask whether the asserted version existed before the critical date and whether its content remained unchanged. Relevant metadata may include whether AI was used, when and how the document was generated, whether a human reviewed it, a cryptographic hash of the original, a trustworthy timestamp, and a version history. Such information should inform evidential reliability without automatically determining legal eligibility.
Strict Single-Reference Disclosure for Novelty
A novelty challenge should identify every claim limitation in one reference, expressly or by necessary implication. It should not reconstruct the invention by retrospectively selecting isolated items from a massive list, combining independent alternatives, or supplying missing essential conditions. An implicit feature must be a necessary teaching of the document as understood by the skilled person.
Graduated Assessment of Enablement and Evidential Weight
Anticipation requires an enabling disclosure. For obviousness or inventive step, a document should be considered only to the extent of what it reliably teaches. Non-existent components, contradictory parameters, missing essential conditions, physical impossibility, and fabricated citations are not a separate statutory “no hallucination” requirement. They are evidence bearing on enablement, motivation to combine, technical plausibility, and reasonable expectation of success.
IV. Human-Centric Filters: Policy Proposals, Not Existing Law
Some scholars argue that ordinary public-accessibility rules may be insufficient in a world of automated, high-volume publication. Their proposals add a link between prior art and human attention, conception, or understanding:
- Human attention: Zhang and Hou propose considering whether a publication received sufficient attention from a human audience, rather than attempting to draw a categorical line between human and AI authorship.
- Conception or recognition: Yordy proposes reintroducing a conception-based requirement into novelty analysis to address the incentive effects of mass-produced AI prior art.
- Substantive nexus to human understanding: Villasenor argues that computer-generated invention descriptions lacking a substantive connection to human understanding should not qualify as printed publications.
Human-centric filters could reduce the effect of low-quality automated publication, but they also create difficulties: actual readership may be accidental and hard to prove; engagement metrics can be manipulated; large platforms may receive preferential treatment; and useful but not yet discovered publications may be excluded. Human attention may therefore work better as corroborating evidence of accessibility and reliability than as a universal threshold requirement.
V. Four Proposals for Mass-Generated AI Prior Art
Require Verifiable Publication Dates and Practical Discoverability
Operators of high-volume defensive-publication repositories should provide stable publication dates, persistent addresses, technical classifications, searchable metadata, and version records. A party relying on a document should identify how the relevant public could have found it before the critical date. Actual readership should remain supporting evidence rather than an automatic prerequisite.
Adopt Integrity Standards for Generation, Review, and Modification History
Publication systems should record whether AI was used, the generation time and method, human review, an original-content hash, a trustworthy timestamp, and subsequent amendments. Missing metadata should not automatically disqualify the reference, but it may weaken proof of the publication date, authenticity, or identity of the asserted version.
Prevent Hindsight Assembly Through Strict Single-Reference Analysis
Where a document lists enormous numbers of components, values, or combinations, decision-makers should ask whether the claimed combination is genuinely individualized in the reference. Novelty should not be denied where the challenger must select multiple items from unrelated lists or add an undisclosed essential condition. Necessary implications and the ordinary use of common general knowledge should remain available.
Separate Novelty from Inventive Step Through Graduated Evidential Weight
Novelty should require an enabling single-reference disclosure. In inventive-step analysis, a non-enabling document should not be categorically excluded, but it should receive weight only for the concrete teaching it reliably conveys. Hallucinated, contradictory, or irreproducible passages should not establish a motivation to combine or a reasonable expectation of success. When a specific and credible enablement challenge is raised, the examiner or challenger should explain why the document and common general knowledge nevertheless support the asserted teaching.
VI. Conclusion
Mass-generated AI documents may increase search costs and legal uncertainty. The available evidence does not, however, establish that they have already undermined the foundations of the patent system. Existing doctrine contains important filters: public accessibility, identification of the actual disclosure, strict single-reference novelty analysis, enablement, and calibrated evidential weight.
The central task is not to assign AI-generated documents an inherently inferior legal status. It is to apply source-neutral rules rigorously in a high-volume environment and to build procedures that make publication date, version, provenance, and review history verifiable. Novelty and inventive step must remain distinct, and an automatically generated document should receive no more evidential weight than the technical teaching it actually and reliably communicates.
References and Official Materials
- 1. USPTO, MPEP § 2121 — Prior Art; General Level of Operability Required to Make a Prima Facie Case Official USPTO material Explains the examination-stage presumption and rebuttal framework, including facial non-enablement arguments under In re Morsa.
- 2. Agilent Technologies, Inc. v. Synthego Corp., Nos. 2023-2186, 2023-2187 (Fed. Cir. June 11, 2025) Federal Circuit opinion (PDF) Addresses anticipatory enablement, prophetic examples, reduction to practice, and proof of efficacy.
- 3. Jazz Pharmaceuticals, Inc. v. Amneal Pharmaceuticals, LLC, 895 F.3d 1347 (Fed. Cir. 2018) Federal Circuit opinion (PDF) Addresses public accessibility of meeting materials and the absence of a universal actual-readership requirement.
- 4. Blue Calypso, LLC v. Groupon, Inc., 815 F.3d 1331 (Fed. Cir. 2016) Opinion Examines searchability, indexing, and posting on a website directed to interested skilled persons.
- 5. UK Intellectual Property Office, Manual of Patent Practice and Examination Guidance for Chemical and Medical Inventions Official UKIPO guidance Summarizes the distinct disclosure and enablement requirements associated with Synthon.
- 6. EPO, Case Law of the Boards of Appeal — Novelty: Implicit Features and Enabling Disclosure Direct and unambiguous derivability · Reproducible disclosure Official summaries of the EPO novelty standards applied in this article.
- 7. Korean Intellectual Property Office, Patent Examination Guidelines Official KIPO materials The current Korean guidelines should be read together with relevant Korean decisions in any specific matter.
- 8. Japan Patent Office, Examination Guidelines and Handbook for Patent and Utility Model JPO Examination Handbook · Internet-disclosure guidance Official materials concerning patentability and technical information disclosed online.
- 9. USPTO, Request for Comments Regarding the Impact of the Proliferation of AI on Prior Art and PHOSITA, 89 Fed. Reg. 34217 (Apr. 30, 2024) USPTO notice The USPTO’s official consultation on AI, prior art, PHOSITA, and patentability determinations.
- 10. Lucas R. Yordy, The Library of Babel for Prior Art, 74 Vanderbilt Law Review 521 (2021) Vanderbilt Law Review repository Proposes a conception-oriented response to mass-produced AI prior art.
- 11. Andrew Zhang & Stephen M. Hou, Attention Is All You Need: Prior Art in the Age of AI (May 7, 2025), SSRN Abstract No. 5245584 SSRN Proposes a human-attention criterion while avoiding a categorical human-versus-AI authorship rule.
- 12. John Villasenor, Ten Thousand AI Systems Typing on Keyboards: Generative AI in Patent Applications and Preemptive Prior Art, 26 Vanderbilt Journal of Entertainment & Technology Law (2024) Vanderbilt JETLaw repository Argues for a substantive nexus to human understanding. The repository currently marks the paper as withdrawn, a status that should be disclosed when citing it.
