AI usage policy
Working version — April 2026
1. Purpose and scope
Scienza in rete is a science communication outlet that bases its authority on the quality of its sources, on interpretive correctness and on direct ties with the Italian scientific community. The introduction of artificial intelligence tools into the editorial cycle is today unavoidable but not neutral: it can improve the efficiency and accessibility of editorial work, but it can also introduce content errors, opacity in processes, attribution problems and risks to sources.
This policy sets out the boundaries within which staff, authors and external collaborators may use generative AI tools, consistent with the duties of scientific journalism, with the European AI Regulation (Regulation EU 2024/1689) and with the trust of the readers of this outlet.
The policy does not regulate AI as a technology, but its use in editorial processes. It applies to all content published under the Scienza in rete byline and to all materials that feed the site, newsletter, social channels and derivative publications.
2. Core principle
Artificial intelligence is a supporting tool. It is not an editorial subject, nor a source, nor an author. Decisions on newsworthiness, reliability of sources, interpretive correctness and final form of content remain always assigned to identifiable persons, accountable both to readers and to the scientific community.
Every piece published by Scienza in rete must be defensible by its author or editor even without the assistance of any system that may have supported it.
3. Applicability
The policy applies to:
- editors and permanent contributors
- external authors (scientists, researchers, journalists) who submit contributions
- fact-checkers, social-media editors, graphic designers, data journalists, videomakers
- anyone working on collateral products: newsletter, podcasts, video, infographics, special issues
It also applies to all phases of the editorial process: preliminary research, translation, transcription, writing, editing, headlining, visual production, data visualization, distribution, archiving and corrections.
The fact that a piece is signed by an external author does not exempt it from the policy: the editorial team is required to verify that external contributions also respect these rules. For this purpose, a usage declaration is requested at submission (see §8).
4. General principles
Substantive human oversight. Every AI-generated output is material to be verified, not content to be trusted by default. Review is not a formality.
Traceability. For editorially relevant uses it must be possible to reconstruct which tool was used, for which function, on which materials and with what level of impact on the final product.
Proportionality. The more a tool affects the content or the form perceived by the reader, the more intense the control must be.
Correct attribution. AI cannot become a way to conceal appropriation of third-party text or to rework copyrighted content without proper citation.
Protection of sources. No confidential document, embargoed material, confidential source, personal data or manuscript under peer review may be uploaded to commercial AI systems without explicit authorization and the use of approved tools.
No delegation of responsibility. AI does not take on editorial responsibility. The byline and the editorial team answer for it in full.
Transparency toward the reader. When the use of AI has a substantive impact on the understanding or perceived authenticity of content, the reader must be informed.
5. Permitted, to-evaluate and prohibited uses
5.1 Generally permitted
Permitted without specific authorization, but always under human control:
- linguistic assistance (grammar correction, local rephrasing, readability adjustments)
- translation of English scientific papers as a reading aid, not as a direct base for the published piece
- transcription of interviews and recordings, as long as the material is not confidential
- internal work organization: meeting summaries, outlines, sorting of materials
- preliminary research: identifying cited papers, definitions, comparing documents, flagging inconsistencies
- suggestions for structure or for chart types given a dataset
- drafts of internal summaries or preparatory briefs
In all these cases no reader disclosure is required, but control remains the author’s or editor’s responsibility.
5.2 Uses that require editorial evaluation
These require explicit discussion with the editor-in-chief or the responsible editor:
- AI-generated drafts of explanatory pieces starting from a preprint or paper (only allowed if the editor verifies every interpretive step against the original)
- construction of an article’s argumentation with AI support
- selection of sources suggested by the system (AI can suggest, but selection is human and documented)
- generation of synthetic illustrations to accompany articles
- data visualizations produced with AI support
- headlines and decks proposed by the system when they affect scientific meaning
- automatic generation of newsletter or social content from the outlet’s material
In these cases the editorial team decides whether reader disclosure is required (§9).
5.3 Uses incompatible with this policy
Prohibited, without exception:
- automatic publication of content without human review
- fabrication or undeclared alteration of citations, data, images or results
- use of AI to conceal appropriation of third-party text
- uploading into commercial AI systems of confidential documents, manuscripts under peer review, embargoed materials, personal data, confidential correspondence with sources
- manipulation of documentary images (lab images, astronomical images, micrographs, clinical images, figures taken from studies) in ways that alter their scientific meaning or simulate results that were not observed
- production of synthetic materials (photo, audio, video) that may be confused with authentic documents without explicit labeling
- use of the system as a substitute for fact-checking
- use of the system as a substitute for reading the primary source when the piece reports a specific scientific result
- uses falling under the prohibited practices in article 5 of Regulation EU 2024/1689
6. Specific cases
6.1 Text
AI may simplify overly technical passages, propose alternative structures, summarize preliminary literature. Any text derived from scientific papers must be re-checked against the original document. Every citation, number, date, drug name and statement on scientific consensus must be verified against a primary source.
6.2 Translation
For Scienza in rete translation is routine: articles and releases in English, later adapted. AI translation as support is allowed, but every technically sensitive passage must be checked. If the published piece is substantially the translation of someone else’s content, the rules on attribution and copyright applicable to a manual translation apply.
6.3 Images
We distinguish between documentary images (lab photographs, astronomical images, micrographs, figures taken from studies), illustrative images (stock photographs, contextual images, decorative) and synthetic images (generated or substantially modified with AI).
For documentary images, AI use is limited to legibility improvements that do not alter meaning. For illustrative images, AI use is permitted but must be disclosed in the caption when the image is synthetic or heavily modified. Synthetic images can never be presented as proof or documents.
6.4 Data visualization
AI may suggest visualization forms, generate draft code, propose axis labels and legends. No chart is published without verification of the dataset, transformation method and correct statistical representation. Every visualization must indicate the data source, the essential processing method and the human responsible for review.
AI cannot be delegated the causal interpretation of data, nor the automated construction of charts without control of denominators, scales and uncertainty.
6.5 Sources and verification
AI can assist discovery: identifying cited papers, searching definitions, comparing tables, flagging points to explore further. It cannot serve as an autonomous cognitive authority. Every scientific claim that affects information for the reader must be traced back to a primary or authoritative secondary source read directly.
6.6 Embargoed materials and peer review
Scienza in rete regularly receives embargoed materials and interacts with the scientific community through channels that entail duties of confidentiality. These materials cannot be uploaded to commercial AI systems. If AI support is needed on this content, only an editorially approved tool that guarantees non-retention and non-training on data may be used (otherwise, work proceeds without AI).
7. Simplified editorial workflow
To avoid burdening production, the workflow concentrates on cases in which AI has a substantive impact on content:
1. Usage declaration. If during production an AI system was used in a non-instrumental way (beyond linguistic correction or transcription), the author or editor notes it in the working file.
2. Risk classification. The responsible editor determines whether the use is low, medium or high editorial risk, according to the criteria in §5.
3. Verification. For medium- or high-risk uses, the verifications in §6 apply to the relevant content type.
4. Disclosure decision. The editorial team decides whether the reader should be informed (§9).
5. Archiving. Minimal notes are kept (tool, purpose, date of use, checks performed) in the piece’s editorial file.
Low-risk uses do not require formal documentation but remain covered by the general principles.
8. External contributions
When submitting an original contribution, external authors are asked to declare if and how they used AI tools in producing the text. The declaration is explicit in the submission email or in the author form.
The editorial team verifies the declared steps. If the editorial review finds signs of undisclosed AI use (inconsistencies in citations, typical errors of generative systems, nonexistent bibliographic references), the piece is discussed with the author before any decision on publication.
9. Transparency toward readers
Scienza in rete adopts a simple disclosure rule:
No disclosure for instrumental AI uses (linguistic correction, transcription, preliminary research, organizational support). These uses do not change the identity of the content.
Disclosure at the bottom of the piece when AI has played a substantive role in drafting, in full translation or in generating visual elements. The standard formula is: “In the production of this article [tool] was used for [function]. The content has been verified and signed by [author/editor].”
Explicit label on synthetic images or materials, even illustrative ones, with the wording “Image generated with AI” or equivalent.
No evasive disclosure. Formulas like “this article may contain AI-generated elements” are not allowed: either we declare precisely, or we do not use AI.
10. Approved tools and personal tools
The editorial team maintains a list of AI tools approved for editorial use, with conditions (data retention, training on input, server location, costs covered by the outlet).
Authors’ personal tools may be used for low-risk purposes. For medium- and high-risk uses, and for any use involving confidential materials, only approved tools may be used.
The list is reviewed at least annually by the editorial team together with the technical lead.
11. Training
Every new contributor receives a copy of this policy and a brief onboarding session on approved tools and typical cases. The editorial team organizes an annual refresher covering: errors encountered, regulatory developments, new tools available.
12. Governance and revision
This policy is signed by the editor-in-chief and approved by the editorial team. It is reviewed annually, or sooner if:
- substantive changes in the European or Italian regulatory framework occur
- editorial incidents involving AI use occur
- substantial changes happen in the landscape of available tools
- Zadig srl, as publisher, issues specific instructions
Any member of the editorial team may propose a motivated revision at any time.
13. Violations
Failure to comply with the policy entails, depending on severity: notice to the editor-in-chief, a request for public correction, suspension of the collaboration, termination of the editorial relationship. For external authors, a violation may result in retraction of the piece or a corrective note appended to it.
14. Short public-facing wording
For the “About” page of the site, the following concise wording may be adopted:
Scienza in rete uses artificial-intelligence tools as a support to editorial work. No content is published without substantive source verification and final human control. Artificial intelligence does not replace the responsibility of the byline and of the editorial team. When AI has a substantive impact on content, we disclose it to the reader. Images and synthetic materials are always labeled.
Document prepared as a basis for editorial discussion. To be reviewed with the editor-in-chief and editorial coordination before adoption.