Data Hostage: How a Preprint Bartered 100,000 Bulk Unreleased Cybersecurity Scenarios for Peer Review Acceptance
Author: Berend Watchus Independent AI & Cybersecurity Researcher [Publication for: OSINT Team]
Automated Generation of Cybersecurity Exercise Scenarios
Data Hostage: How a Preprint Bartered 100,000 Bulk Unreleased Cybersecurity Scenarios for Peer Review Acceptance
A structural integrity analysis of arXiv:2604.01079
This is what you see.
You find this on arXiv, April 1, 2026:
Automated Generation of Cybersecurity Exercise Scenarios Charilaos Skandylas, Mikael Asplund — Linköping University
The abstract closes with this sentence:
“We further release as open source: a simulation and a virtualization environment that can run cybersecurity exercises based on the generated scenarios and a dataset containing 100000 sample scenarios.”
Present tense. Unqualified. Definitive.
You would reasonably conclude: the scenarios exist, they are released, they are accessible now.
None of that is true.
This is what they admit.
Buried at the very end of the paper, in Appendix B titled — with notable irony — “Open Science,” a single paragraph reads:
“In the interest of open science, we will provide our dataset alongside a snapshot of the scenario generator, simulator and virtualization sources, released under an open source license upon acceptance.”
Read that carefully against the abstract.
The abstract says: we release. The appendix says: we will release if you accept us.
That gap is not a writing error. It is the entire mechanism.
But it is actually worse than that.
The Number Disappears
The abstract promises something specific: 100,000 sample scenarios. A concrete, countable, verifiable quantity. Large enough to impress. Specific enough to suggest rigorous generation and validation.
Now look at what Appendix B actually commits to releasing:
“our dataset.”
That is all. No number. No quantity. No specification. No verification commitment. Just “our dataset” — an undefined, unquantified, unspecified blob of future maybe.
The 100,000 figure exists in exactly one place in this paper: the abstract, where it does maximum persuasive work on every reader and every peer reviewer who will ever encounter this paper. By the time the authors commit to an actual release in Appendix B, the number has silently vanished.
This means that upon acceptance — if acceptance comes — the authors could release 12 scenarios, a partial generator snapshot, and a readme file, and have technically honored every commitment made in Appendix B.
The 100,000 is a promise made where it counts and erased where it binds.
The Transaction
Those two statements together — the abstract and Appendix B — form an explicit barter that no legitimate open science framework would recognize.
The dataset is not coming eventually. It is not “maybe later.” It has a specific, named release trigger: peer review acceptance. That trigger is a direct transaction:
- Give us acceptance
- We give you something we will call “our dataset”
The 100,000 scenarios — which by academic convention should accompany the paper as a verified, accessible dataset at submission — are instead being held as transaction currency. Peer reviewers are being asked to validate a contribution they cannot see, in exchange for a release that only happens if validation succeeds, and which carries no commitment to the quantity that made it sound valuable in the first place.
That is not open science. That is a hostage structure. The abstract is the ‘ransom’ note. Appendix B is the small print.
The Word Doing the Dishonest Work
In academic abstract convention, “we further release” means: in addition to the methodology we just described, we additionally provide these concrete deliverables — now, accompanying this submission.
It is the same register as “source code available at github.com/X” or “dataset available at DOI:Y.” It is a statement of present fact, not future intent.
A legitimate future intent statement would read: “We will release upon acceptance” or “We plan to make available.” Those phrasings appear nowhere in the abstract. Only in Appendix B, which most readers and many reviewers will never reach, does the condition surface — and by then the number has already been quietly dropped.
The word “further” makes it worse. “We further release” positions the dataset as a co-equal contribution alongside the paper itself. It is the payoff sentence of the abstract. It is the answer to “so what do we actually get?”
The answer, in Appendix B: something unspecified, sometime after acceptance.
What Peer Reviewers Are Actually Being Asked to Do
When stated plainly, the situation is extraordinary:
- Evaluate a simulation environment they cannot access
- Evaluate a virtualization environment they cannot access
- Evaluate 100,000 scenarios they cannot access — and which the authors have not even committed to releasing in that quantity
- Verify empirical claims about scalability, variety and difficulty that rest entirely on data they cannot access
- Grant acceptance, after which something called “our dataset” will be released
This inverts the entire logic of peer review. Peer review exists to verify claims before they receive academic endorsement. Here verification is structurally impossible because the evidence is withheld until after endorsement is granted — and the quantity of evidence promised in the abstract is not even guaranteed in the release commitment.
What 100,000 Scenarios Actually Are
Even setting aside the barter structure, the 100,000 scenarios deserve scrutiny as a headline contribution.
They are not hand-crafted. They are not expert-validated. They are not curated. They are the automated output of an Alloy model finder — a formal constraint satisfaction tool — running against specifications the authors wrote. The number 100,000 is a parameter. They could have set it to 1,000,000 with identical effort. The number communicates scale, not work.
Here is what generating 100,000 scenarios actually involves:
The generation process: Write the Alloy specifications — the real intellectual work, already fully published in the paper text. Configure the generator. Run it. Wait. The paper’s own Table 1 reveals the scale: 125 networks generate in approximately 17 minutes. Scaling to 100,000 is a matter of compute time, not human effort.
The compute budget: A standard cloud instance running this workload costs between €5 and €50. This is not a major infrastructure contribution. It is an overnight batch job.
The human effort after the specifications are written: Run a script.
What 100,000 scenarios look like physically: Each scenario is a structured data object — a graph of network topology, subsystem components, interfaces, vulnerabilities, locks, keys, and objectives. In JSON format, a single scenario occupies roughly 10 to 50KB. At 20KB average, 100,000 scenarios is approximately 2GB. That fits in a single zip file. It would be hosted on Zenodo or a GitHub release in minutes. There is no technical barrier to releasing it today, or the day the paper was submitted.
What 100,000 scenarios are not: They are not 100,000 stories. They are not 100,000 hand-designed training exercises. They are structured data objects that formally satisfy constraint specifications. A human reading one would encounter: network topology type, connector configuration, subsystem components, interface assignments, vulnerability mappings, lock placement, key location, objective target. Multiply that by 100,000. It is a database, not a library. It is bulk output, not curated knowledge.
Anyone Can Do This
The deeper problem with holding this dataset hostage is that it is not scarce.
A non-expert with access to any current commercial LLM and a structured approach could produce a semantically equivalent or superior dataset for under €100 per month. The process is straightforward:
Develop 35 to 100 domain questions covering what constitutes a good cybersecurity exercise scenario. What attack patterns appear in real enterprise environments? What difficulty gradations exist in practice? What makes a scenario useful for training a human analyst versus an AI agent? What network topologies reflect actual enterprise architectures? Use those answers to construct a generation framework. Generate scenarios in bulk. Use the LLM to validate each scenario against established cybersecurity benchmarks.
This runs over a weekend. It costs one monthly LLM subscription. It produces semantically grounded scenarios — reflecting real-world attack patterns and training utility — rather than formally valid but semantically unvalidated constraint satisfactions.
The result is a zip file on Zenodo. Released the same day it was generated because there is no barter to execute, no hostage to hold, no acceptance to wait for.
The Quality That Was Never Checked
There is a question the paper never asks: are these good scenarios?
Formally valid is not semantically useful. A scenario can satisfy every Alloy constraint the authors specified and still be useless for cybersecurity training. The paper’s evaluation measures structural variety via NetSimile graph similarity metrics and content variety via clustering heatmaps. Both measure diversity of output. Neither measures quality of output.
The paper does not ask:
- Do these scenarios reflect attack patterns observed in real enterprise incidents?
- Would a cybersecurity trainer recognize these as meaningful exercises?
- Do the difficulty gradations correlate with real-world operational complexity?
- Would an AI agent trained on these scenarios perform better in actual defensive tasks?
- Has any domain expert reviewed a single scenario for realism?
None of these questions are answered because none were asked. The scenarios are validated against the authors’ own formal model — a closed loop where the generator and the validator share the same assumptions. This is not independent quality validation. It is confirmation that the output matches the specification that produced it.
The ‘ransom’ being demanded is for a dataset whose quantity is unguaranteed, whose quality has never been independently assessed, whose utility has never been demonstrated, and whose generation required minimal human effort beyond the methodology already published in the paper text.
The Pattern and the Warning
This paper exhibits a structure that deserves to be named clearly:
- Abstract makes present-tense open science claim with a specific impressive number
- Appendix reveals the release is conditional on peer review acceptance
- Appendix quietly drops the specific number — committing only to “our dataset”
- Peer reviewers cannot access or verify any of the claimed contributions
- The contribution is trivially replicable by anyone with basic tools and a small budget
- Semantic quality of the contribution is never independently validated
- Release trigger is acceptance — a direct transaction, not a timeline
- The quantity promised in the abstract is not guaranteed in the release commitment
The warning for the research community is simple and actionable:
When an abstract uses present tense to describe open science releases, verify that the release actually exists before treating it as a contribution. Search for the repository. Search for the DOI. Search for the dataset. If it does not exist at submission time, the abstract has misrepresented the paper’s state of completion.
Then read the appendix. Check whether the quantity promised in the abstract is actually committed to in the release terms. If the number has disappeared, you are looking at a two-stage deception: impress with the abstract, bind with the appendix, deliver whatever is convenient after acceptance.
Peer review cannot function correctly when the evidence under review is withheld pending the outcome of the review. That is not a procedural inconvenience. It is a structural corruption of the verification process that peer review exists to perform.
The data is not released. The scenarios are bulk automated output. The quality is unvalidated. The quantity is unguaranteed. The transaction is explicit.
The ‘ransom note’ is the abstract. The small print is Appendix B.
arXiv:2604.01079 — Skandylas & Asplund — Linköping University Submitted: April 1, 2026 Abstract promises: 100,000 scenarios Appendix B commits to: “our dataset” Dataset status: not released Simulation environment status: not released Virtualization environment status: not released
Release condition: peer review acceptance
— — — — — — — — — — — — — —
EDIT1:
— — — EDIT [April 2, 2026] — — —
The Endorsement Layer
One structural detail about this paper’s presence on arXiv deserves explicit documentation.
arXiv does not accept open submissions. To publish a preprint there, a new submitter requires endorsement from an established arXiv member — someone with a prior publication record in the relevant subject area who explicitly vouches for the submitter’s legitimacy. The endorsement is a reputational transfer. The endorser stakes their credibility on the submission being a good-faith scientific contribution.
That endorsement system is arXiv’s primary quality gate.
What it filters: the legitimacy of the submitter.
What it does not filter: the accuracy of the abstract.
That gap is precisely where the manipulation in arXiv:2604.01079 operates. Someone vouched for these authors. That vouching gave the submission institutional credibility — Linköping University affiliation plus an endorsement chain — which is then read by downstream systems, including Google AI, as a signal that the content has passed a credibility filter. Google AI summarized the abstract as established fact within hours of submission, in part because the submission arrived pre-legitimized.
The endorsement system, designed to prevent low-quality submissions, functioned here as an unintentional trust amplifier for a misleading abstract. The vouching did not protect against abstract-level misrepresentation. It accelerated its propagation.
The person who endorsed this submission vouched for the researchers. They did not — and could not — vouch for the claim that 100,000 scenarios exist and are released right now. But the downstream ecosystem cannot make that distinction. It reads the endorsement as applying to the entire submission.
That is a structural vulnerability in the arXiv endorsement model that this case makes visible for the first time in documented form.
— — — END EDIT — — —
— — — — — — — — — — — — — —
archives
— — — —
— — — — — — — — —
Google ai mode
Dit is een scherpzinnige observatie die de kern raakt van de huidige debatten over Open Science en de integriteit van het preprint-systeem.
Hieronder volgt de uitleg in het Engels, Chinees en Spaans over de ethische implicaties van “voorwaardelijke data-vrijgave” in ruil voor peer review.
English: The Ethics of Conditional Data Release
When authors claim to have a dataset but withhold it behind a “transactional barrier” (releasing it only upon acceptance of the paper), it creates several academic and ethical issues:
- Peer Review vs. Open Science: Peer review is a quality control mechanism, not a bargaining chip. By withholding data until acceptance, authors prevent the community from verifying their claims before the paper becomes “official.” This contradicts the fundamental principle of Open Science: transparency.
- The “Hostage” Problem: As your source suggests, this is a form of “data hostage-taking.” It forces reviewers to trust the results without seeing the evidence, essentially asking for a leap of faith rather than a scientific validation.
- Devaluation of Preprints: Preprints on platforms like arXiv are intended to accelerate the sharing of knowledge. If a preprint makes grand claims (like “100,000 networks”) that cannot be verified because the data is hidden, the preprint loses its scientific utility and becomes a mere advertisement or a “claim-staking” exercise. This erodes trust in arXiv as a source of credible, early-stage research.
Chinese (中文): 条件性数据发布的伦理问题
如果一篇论文承诺提供数据集，但将其作为通过同行评审的“交换条件”，这在学术界引发了严重的诚信质疑：
- 学术诚信与透明度： 开放科学的核心在于可重复性。如果数据集在论文被接受之前不可用，其他研究人员就无法验证作者的实验结果。这种“先接受再发布”的做法违背了科学透明的原则。
- 同行评审的贬值： 同行评审本应基于对数据和方法的严格审查。如果数据被扣为人质，评审员就无法履行职责，导致评审过程流于形式。
- 对 arXiv 价值的影响： arXiv 的初衷是促进知识的快速传播。如果预印本变成了一种“营销工具”，发布无法验证的海量数据声明（如 100,000 个网络），它就会损害预印本平台的声誉，使其从科学交流平台降级为虚假宣传的温床。
Spanish: La ética de la liberación condicional de datos
Cuando un artículo promete un conjunto de datos pero lo utiliza como una moneda de cambio para la aceptación en revisión por pares, se plantean serios problemas éticos:
- Ciencia Abierta vs. Transacción: La integridad académica exige que los datos que respaldan las conclusiones estén disponibles para su verificación. Condicionar la liberación de los datos a la aceptación del artículo es una práctica transaccional que obstruye la reproducibilidad.
- El Secuestro de Datos: Como menciona su fuente, esto actúa como un “secuestro de datos”. Se pide a los revisores y a la comunidad científica que validen un trabajo cuyas pruebas fundamentales están ocultas. Esto invalida el propósito del control de calidad científico.
- Impacto en arXiv: arXiv es un pilar de la comunicación científica rápida. Si los autores lo utilizan para publicar resultados que no se pueden verificar, la plataforma se llena de “promesas” en lugar de ciencia real. Esto debilita la confianza en los preprints y perjudica a los investigadores que sí practican una transparencia total.
Conclusie (Nederlands)
Het aanbieden van een dataset als “beloning” voor acceptatie (in plaats van als bewijslast voor de review) wordt in de moderne wetenschap vaak gezien als een slechte praktijk (bad practice). Het ondermijnt het proces van falsificatie: als men de data niet kan inzien, kan men de fouten niet vinden. Voor arXiv betekent dit een risico op “vervuiling” met claims die wel de aandacht trekken, maar niet de wetenschappelijke toets der kritiek kunnen doorstaan.
Russian (Русский): Этика условного выпуска данных
Когда авторы обещают предоставить набор данных (dataset) только после принятия статьи рецензентами, это создает серьезные этические и академические проблемы:
- Открытая наука против сделки: Суть открытой науки заключается в прозрачности. Если данные скрыты до момента публикации, научное сообщество не может проверить утверждения авторов. Это превращает научный процесс в транзакцию, а не в поиск истины.
- Проблема «заложников данных»: Как указывают критики, это форма «захвата данных». Рецензентов просят подтвердить качество работы, не давая им доступа к доказательствам. Это подрывает саму суть рецензирования.
- Обесценивание arXiv: Препринты предназначены для ускорения обмена знаниями. Если препринт содержит громкие заявления (например, о «100 000 сетях»), которые невозможно проверить, arXiv превращается из научной платформы в рекламную площадку, что подрывает доверие к предварительным публикациям.
Portuguese (Português): A Ética da Liberação Condicional de Dados
Quando os autores prometem um conjunto de dados, mas o mantêm sob uma “barreira transacional” (liberando-o apenas após a aceitação do artigo), surgem vários problemas éticos:
- Ciência Aberta vs. Transação: A integridade acadêmica exige que os dados que sustentam as conclusões estejam disponíveis para verificação. Condicionar a liberação à aceitação do artigo obstrui a reprodutibilidade e viola os princípios da Ciência Aberta.
- O Problema do “Refém de Dados”: Como sugere a sua fonte, trata-se de uma forma de manter os dados como reféns. Obriga os revisores a confiar nos resultados sem ver as evidências, transformando a validação científica em um ato de fé.
- Desvalorização de Preprints: O arXiv serve para acelerar o compartilhamento de conhecimento. Se um preprint faz grandes alegações que não podem ser verificadas porque os dados estão ocultos, a plataforma perde sua utilidade científica e torna-se apenas um meio de “marcar território” sem prova.
Japanese (日本語): データ公開の条件付き取引に関する倫理
論文の著者がデータセットの存在を主張しながら、「論文が採択された場合のみ公開する」という条件を付けることは、学術的・倫理的に大きな問題があります。
- オープンサイエンスか、取引か： オープンサイエンスの根幹は透明性です。採択までデータを非公開にすることは、コミュニティによる検証を妨げるものであり、科学的な透明性の原則に反します。
- 「データの質入れ」問題： 指摘されている通り、これはデータを「人質」に取る行為です。査読者は証拠を確認できないまま結果を信頼せざるを得ず、科学的な検証が「憶測」に取って代わられてしまいます。
- arXiv の価値の低下： arXiv（プレプリント）の目的は、知識の迅速な共有です。検証不可能な大規模な主張（例：10万個のネットワーク）を掲載しながらデータを隠すことは、arXiv を科学の場から単なる「宣伝の場」へと貶め、プレプリント全体の信頼性を損なうことになります。
Korte samenvatting (NL)
Deze praktijk verandert peer review van een inspectie in een gijzeling. Het ondermijnt de geloofwaardigheid van preprints omdat de geclaimde waarde (de 100.000 scenario’s) niet geverifieerd kan worden op het moment dat de wetenschappelijke claim wordt gemaakt.
— — — — — — — — — —
another google ai mode session says:
English Correction
Based on the verbatim text of arXiv:2604.01079, there is a direct contradiction regarding the availability of the materials. The correct status as of April 2, 2026, is as follows:
- No Current Availability: Despite the abstract using present tense (“we further release”), Appendix B (Open Science) explicitly states that the dataset, simulator, and virtualization sources will only be provided “upon acceptance” of the paper.
- Contradictory Claims: There is a discrepancy between the abstract, which promises 100,000 scenarios, and Appendix B, which only commits to “our dataset” without confirming that specific quantity.
- Verification: There are currently no public URLs, DOIs, GitHub repositories, or Zenodo entries for these materials. The claim of a current release is factually incorrect; it is a conditional future release.
中文更正 (Chinese)
根据 arXiv:2604.01079 论文原文，关于该研究资料的开放状态存在明显的矛盾。截至 2026 年 4 月 2 日，正确情况如下：
- 当前不可获得：尽管摘要（Abstract）使用了现在时态（”we further release”），但论文的 附录 B (Open Science) 明确指出，数据集、模拟器和虚拟化源码将在论文被接收后 (upon acceptance) 才会发布。
- 内容表述不一：摘要中明确承诺提供 100,000 个场景，但在附录 B 的正式承诺中，仅提到提供“我们的数据集”，并未再次确认具体的数量级。
- 核实结果：目前不存在任何公开的 URL、DOI、GitHub 仓库或 Zenodo 条目。之前的发布声明在事实上是不准确的；这属于有条件的未来发布，而非现有的资源。
Русский (Russisch)
На основании текста документа arXiv:2604.01079 существует явное противоречие относительно доступности материалов. Корректный статус на 2 апреля 2026 года:
- Отсутствие доступности: Несмотря на использование в аннотации настоящего времени («мы выпускаем»), в Приложении B (Open Science) прямо указано, что набор данных и исходные коды будут предоставлены только «после принятия» (upon acceptance) статьи.
- Противоречивые утверждения: Существует расхождение между аннотацией, обещающей 100 000 сценариев, и Приложением B, которое упоминает лишь «наш набор данных» без подтверждения этого количества.
- Проверка: На данный момент не существует публичных URL, DOI, репозиториев GitHub или записей Zenodo для этих материалов. Утверждение о текущем выпуске фактически неверно; это условный выпуск в будущем.
Português (Portugees)
Com base no texto de arXiv:2604.01079, há uma contradição direta sobre a disponibilidade dos materiais. O status correto em 2 de abril de 2026 é:
- Indisponibilidade Atual: Embora o resumo utilize o presente do indicativo (“nós lançamos”), o Apêndice B (Open Science) afirma explicitamente que o conjunto de dados e os códigos-fonte serão fornecidos apenas “após a aceitação” (upon acceptance) do artigo.
- Alegações Contraditórias: Há uma discrepância entre o resumo, que promete 100.000 cenários, e o Apêndice B, que se compromete apenas com “nosso conjunto de dados”, sem confirmar essa quantidade específica.
- Verificação: Atualmente não existem URLs públicas, DOIs, repositórios GitHub ou entradas no Zenodo para estes materiais. A afirmação de um lançamento atual está factualmente incorreta; trata-se de um lançamento futuro condicional.
Español (Spaans)
Según el texto de arXiv:2604.01079, existe una contradicción directa sobre la disponibilidad de los materiales. El estado correcto a fecha de 2 de abril de 2026 es:
- Sin disponibilidad actual: Aunque el resumen utiliza el tiempo presente (“lanzamos”), el Apéndice B (Open Science) indica explícitamente que el conjunto de datos y los códigos fuente se proporcionarán solo “tras la aceptación” (upon acceptance) del artículo.
- Afirmaciones contradictorias: Hay una discrepancia entre el resumen, que promete 100.000 escenarios, y el Apéndice B, que solo se compromete con “nuestro conjunto de datos” sin confirmar esa cantidad específica.
- Verificación: Actualmente no existen URL públicas, DOI, repositorios de GitHub ni entradas en Zenodo para estos materiales. La afirmación de un lanzamiento actual es tácticamente incorrecta; es un lanzamiento futuro condicional.
日本語 (Japans)
論文 arXiv:2604.01079 の本文に基づくと、資料の利用可能性に関して直接的な矛盾があります。2026年4月2日時点の正確な状況は以下の通りです。
- 現在は利用不可: 要約（Abstract）では現在形（「公開する」）が使われていますが、付録B (Open Science) には、データセットおよびソースコードは論文が**「採択された時点で (upon acceptance)」**提供されると明記されています。
- 矛盾する主張: 要約では 10万件のシナリオ を約束していますが、付録Bでは具体的な数量には触れず、「我々のデータセット」と述べるにとどまっています。
- 事実確認: 現在、これらの資料に関する公開URL、DOI、GitHubリポジトリ、Zenodoのエントリは存在しません。現時点で公開済みという主張は事実誤認であり、正しくは**「条件付きの将来的な公開」**です。
Data Hostage: How a Preprint Bartered 100,000 Bulk Unreleased Cybersecurity Scenarios for Peer… was originally published in OSINT Team on Medium, where people are continuing the conversation by highlighting and responding to this story.