AI and Data Usage Disclosure

Preliminary

P.1 · How XSci uses AI and data. XSci uses artificial intelligence in certain services to support manuscript analysis, report generation, and AI-assisted human review where applicable. This page explains, in practical terms, how AI may be used on the platform, what role it plays, what limitations apply, and what XSci does with the manuscripts, reports and review activity that pass through it. It turns on a single distinction: whether you have made something public. Personal and account data — who you are, how you log in, how you are billed — is governed by the Privacy Policy and not by this page.

P.2 · No machine certifies anything on XSci. One thing is fixed and appears first because everything else depends on it. Certification is a judgement made by a compensated, named human expert under the Certification Standard, and no AI system can produce it — alone, in combination with another system, or by any degree of assistance. What AI can do on XSci is produce an AI review report — a written opinion on a manuscript, and not a certification.

P.3 · Definitions. These terms are used in the same sense here, in the Certification Standard, the Peer Review Policy and the Reviewer Agreement.

• AI review report — a written opinion on a manuscript produced by an AI system, presented as a narrative report.
• Review system — the models, prompts, tooling and processes by which XSci produces AI review reports.
• Train and fine-tune — to adjust the parameters of a model using material, so that the material shapes what the model produces afterwards.
• Evaluate — to measure how well a review system performs, by comparing its outputs against material or outcomes, without adjusting the model's parameters.
• Public deposit — a manuscript version you have deposited publicly on the platform, with its DOI and its record.
• Public artefact — anything published on the platform and visible to anyone: a public deposit, a released report, a certification record, an author response.
• Private interaction data — what you do on the platform that is not published: the record of your actions, and the text you write about a review. It does not include your manuscript or the materials you supply with it.
• Confidential material — the manuscript and anything else you supply in order to obtain a service and do not publish, including underlying materials provided privately to a reviewer. It does not include private interaction data, which is a separate class.
• Provider — a company, including XSci itself, that supplies AI review on the platform.
• Certification decision — the binary judgement of a named human expert under the Certification Standard.

1. Where AI may be used

1.1 · What XSci may use AI to do. Depending on the service model and workflow:

(a) analyze manuscript content; (b) generate structured reports or feedback; (c) support review-related assessments and recommendations; (d) assist with consistency, formatting, or presentation of outputs.

1.2 · The role of AI may differ across XSci services. In some cases, AI-generated outputs may be provided on their own. In other cases, they may be provided alongside human review or other platform processes.

1.3 · AI review is a marketplace, and you choose the provider. AI review on XSci is offered from more than one provider, and the author selects which to use. The price and description of each are shown on the same terms. Neutrality here means the three things in 1.4 to 1.6.

1.4 · No editorial preference. XSci does not present its own service as the default, recommend it over another, or order the providers to its own advantage.

1.5 · The same commercial terms. Every provider is offered a place on the same terms, and XSci's own service is carried on those terms too.

1.6 · Any quality measure is computed the same way for everyone. Where XSci publishes a measure of a provider's quality, the method is published and applied identically to every provider, including XSci.

1.7 · Selecting a provider. Where you select a provider, your manuscript is sent to that provider to produce your report, on the data terms in clause 4.

1.8 · Free AI review. Where AI review is provided at no charge — as a platform baseline, or to every entry in a competition XSci runs — the marketplace obligations in 1.3 to 1.6 apply to paid AI review, and the free service is what XSci has chosen to provide. The labelling rules in clause 3 apply to it in full.

1.9 · Where a human expert reviews your manuscript, the expert chooses whether to see any AI review report that exists for it. That choice, and the expert's own use of AI in preparing their report, are recorded on the assessment record. An AI review report does not bind a human expert's assessment in any respect.

2. What AI does not do by itself

2.1 · AI does not by itself determine copyright ownership, legal rights, or regulatory compliance.

2.2 · AI does not certify. No certification decision is made by, or delegated to, an AI system. An AI review report may inform the human expert who makes the decision, and the expert decides what weight to give it; it is not a certification, and nothing produced by AI review may carry the certification mark. Unless expressly stated otherwise for a particular service, AI outputs should not be understood as final or authoritative human judgments, and they should not be treated as guarantees of publication, certification, acceptance, or legal accuracy.

2.3 · An AI review report is not a verification. It is a generated narrative report — a written opinion produced by a machine, and not a mechanical test of anything. It is never displayed as though it had established a fact, and it asserts nothing about the adequacy of a manuscript.

2.4 · XSci does not train on your manuscript, or on anything you supply with it. Clause 4 says exactly what that covers: what XSci does with public artefacts, what it does with private interaction data, and what it never does with confidential material. It is a promise about XSci's own products and services, and it does not extend to a third-party provider you choose to use.

3. Transparency and labelling

3.1 · Where applicable, XSci will clearly indicate when AI-generated outputs, AI-assisted analysis, or AI-supported services are involved.

3.2 · Distinguishing AI from human output. Where different types of outputs are presented together — for example, AI-generated material and human-generated material — XSci distinguishes them in a way that is reasonably clear to users. An AI review report is labelled as AI-generated. A report for which a named human expert is accountable is not so labelled, and the expert may have used AI tools as an aid in drafting it; the commitment we make is accountability, not the absence of AI.

3.3 · The AI-reviewed label. A manuscript that has been through AI review may carry the AI-reviewed label on its record. The three rules in 3.4 to 3.6 govern it.

3.4 · It attaches only once you have deposited a revision following the AI review. AI review run with no revision deposited changes nothing on the record.

3.5 · It records that a service was run, and asserts nothing about the adequacy of that service or the quality of the work. It is not a certification and is not a weaker form of one.

3.6 · AI-reviewed may never be shortened to "reviewed" in interface text, marketing, documentation, or any badge image, and its typographic treatment must be such that the qualifier cannot be cropped away.

3.7 · Paid and free alike. The rules in 3.3 to 3.6 apply whether the AI review was paid for or provided free, including in a competition XSci runs. Nothing about the label changes because the service cost nothing.

3.8 · The certification mark is separate. XSci Peer-Certified Preprint is never applied to AI review output and is governed by the Peer Review Policy and the Certification Standard.

4. Data handling

4.1 · One question decides the class. We treat your manuscript as your intellectual property. This clause says what XSci does with three classes of material, and the class a thing falls into is decided by one question: have you made it public?

4.2 Public artefacts

4.2.1 · What they are. A public deposit, a released report, a certification record and an author response are public: anyone can read them, and XSci publishes them because that is what the platform is for.

4.2.2 · What XSci does with them. XSci uses public artefacts:

(a) to provide the services you have asked for; (b) to operate and secure the platform; and (c) to evaluate, research and improve review systems, including by training and fine-tuning XSci's own models.

4.2.3 · The basis for it. The use in 4.2.2 is under the platform licence you grant in the Terms of Service, and it is a consequence of publication rather than a separate permission. There is no opt-out from it beyond the scope of that licence, because the material is already public and the licence is the basis on which XSci publishes it.

4.3 Private interaction data

4.3.1 · What it is. Private interaction data is what you do on the platform that is not published: which findings in a review you accepted, dismissed or changed, how far you revised, what you responded to and when — and the comments, notes and edits you wrote about a review.

4.3.2 · What XSci does with it. XSci may use it, de-identified, to evaluate, research and improve review systems, including by training and fine-tuning XSci's own models.

4.3.3 · Why we use it. How a reader uses a review is what tells us whether the review was any good — which findings were acted on, which were dismissed, what a revision actually changed. That evidence is the main thing that makes the next review better, and it cannot be obtained any other way.

4.3.4 · Your manuscript does not enter training through what you write. People quote their own work when commenting on a review. Before interaction text is used to train or fine-tune anything, XSci takes reasonable technical steps to detect and remove passages reproduced from a manuscript, and any passage so identified is treated as confidential material and excluded. The undertaking in 4.4 is not qualified by this clause: material you supply in order to obtain a service is never used to train anything, whether you supplied it as a file or quoted it in a comment.

4.3.5 · What "de-identified" means here, and what it does not. We remove the identifiers that tie the data to you directly. We do not claim it is anonymous: de-identified data is still personal data, this Disclosure and the Privacy Policy both apply to it in full, and your rights over it are unaffected. We use the word we can stand behind rather than the word that sounds better.

4.3.6 · Your opt-out, and exactly what it does. You may opt out by contacting privacy@xsci.com or from your account settings. From the moment you do, your interaction data stops being used for these purposes and is excluded from every training run thereafter. What an opt-out cannot do is reach into a model that has already been built: a trained model cannot be untrained, and we will not tell you otherwise. Opting out does not affect any service you have bought, your standing on the platform, or any certification you hold.

4.3.7 · How long it is kept. Interaction data held for these purposes is retained until six years after your last interaction with the platform — a deposit, an assessment, a response, or any other use of the Service — after which it is deleted or irreversibly aggregated. The period matches the one we apply to account and contract records.

4.4 Confidential material

4.4.1 · What it is, and what is done with it. The manuscript and anything else you supply in order to obtain a service and do not publish — including underlying materials provided privately to a reviewer — is used only to deliver the service you bought. It is never used to train, fine-tune or evaluate any model or review system, by XSci or by anyone else. This undertaking has no exceptions, and it holds however the material reaches us: a passage of your manuscript quoted inside a comment is still your manuscript, and 4.3.4 says what we do about it. Private interaction data is not confidential material: it is a separate class, dealt with in 4.3.

4.5 The limit of these promises

4.5.1 · What that promise covers, and what it does not. The commitments in 4.2 to 4.4 are given by XSci and apply to XSci's own products and services. They are not, and cannot be, promises about how an independent company handles your manuscript.

4.5.2 · Third-party AI review services on our platform. AI review on XSci is offered from more than one provider, and where you select a provider other than XSci, your manuscript is sent to that provider and is then handled on that provider's terms, not ours. Those terms are the provider's own, and XSci's undertakings about training do not bind them. We ask you to consult the third-party provider's own policies on data use, model training and retention before you buy their service, and we link to them at the point of purchase.

4.5.3 · What we do not represent. We select the providers we make available with care, and where we can obtain contractual protections for your content we do. We do not represent that any third-party provider is contractually barred from training on your content, and you should not assume it.

5. How we improve our review systems

5.1 · Evaluation against outcomes. XSci evaluates its review systems against outcomes — what a system said about a manuscript, set against what happened to that manuscript afterwards. That evaluation uses public artefacts and de-identified private interaction data, and never confidential material. Improvement may include training and fine-tuning XSci's own models on either, subject to 4.3.4.

5.2 · Expert reports. Reports written by compensated human experts are used to evaluate and improve review systems under the Reviewer Agreement, as a condition of paid engagement. Reviewers are not publicly attributed, which is a longstanding principle of peer review, and evaluation proceeds on an anonymised basis. That anonymity concerns public attribution, not internal evaluation use.

5.3 · Research in the open. XSci may publish aggregate findings about how well review systems perform. Any dataset XSci releases contains public artefacts only.

5.4 · Opting out. You may opt out of the use of your private interaction data at any time by contacting privacy@xsci.com. Opting out does not affect any service you have bought, your standing on the platform, or any certification you hold. There is no opt-out from the use of public artefacts beyond the scope of the platform licence in the Terms of Service, and no opt-in is possible for confidential material, which is never used this way at all.

6. Limitations of AI outputs

6.1 · AI-generated outputs may be useful, but they have limitations. For example:

(a) AI may not fully understand highly specialized or novel research; (b) AI may produce incomplete, inaccurate, or irrelevant observations; (c) AI cannot independently verify all factual claims, experimental results, or the integrity of underlying materials; (d) AI outputs may reflect limitations in training data, model design, or prompt context.

6.2 · AI cannot make the judgements certification requires. Certification turns on whether a manuscript contains a major error and on whether it has a genuine prospect of publication in the venue the authors are aiming at. Both are expert judgements about a specific manuscript in a specific field, and neither is delegable to an AI system.

6.3 · How AI outputs should be understood. For these reasons, as informational or supportive in nature, and not as the sole basis for important academic, editorial, legal, or commercial decisions.

7. Our commitment

7.1 · XSci's approach to AI and data use is guided by the following principles:

(a) No machine certifies. Certification is always a named human expert's judgement, and no certification decision is delegated to an AI system. (b) Neutrality in the marketplace. No editorial preference, the same commercial terms for every provider, and any published quality measure computed by a published method applied identically to all. (c) Honest labelling. AI-generated output is labelled as such; AI-reviewed is never shortened to "reviewed"; and no label asserts more than the fact that a service was run. (d) No training on your manuscript, or on anything you supply with it. Public artefacts are used under the platform licence in the Terms of Service; private interaction data is used de-identified, may be used for training, and you may opt out; confidential material is never used at all, including where you have quoted it yourself. (e) The third-party limit is stated, not hidden. Where you choose a third-party AI review service on our platform, that provider's own policies govern, and we tell you who they are, what model they run, whether they train, and where to read their terms. (f) transparency about the role of AI in relevant services; (g) continued evaluation and improvement of AI-supported services; (h) human expert accountability wherever a report or a decision carries our name.

8. Questions and changes

8.1 · Questions. If you have questions about how XSci uses AI and data, contact us at support@xsci.com.

8.2 · Changes to this page apply prospectively. Where a future version widens what XSci may do with material, it applies to material received after that version takes effect. It does not reach back to material XSci already holds under an earlier version of this page.