AI and Research Integrity Policy | Health Nexus: Digital Health and Medical AI

Health Nexus: Digital Health and Medical AI supports the responsible use, development, evaluation, and reporting of artificial intelligence in scholarly research and publishing. Artificial intelligence may provide substantial benefits in research, clinical care, digital health, data analysis, scientific communication, and editorial production; however, its use also creates risks involving accuracy, bias, transparency, privacy, confidentiality, intellectual property, reproducibility, accountability, and patient safety.

This policy establishes requirements for:

The use of artificial intelligence and AI-assisted technologies by authors;
Research involving artificial intelligence, machine learning, large language models, clinical decision-support systems, digital-health platforms, and related technologies;
The use of AI by reviewers, editors, editorial board members, and the publisher;
Disclosure, documentation, validation, data governance, authorship, accountability, and research integrity;
The investigation and correction of inappropriate, misleading, undisclosed, or unethical AI use.

Human authors, reviewers, and editors remain fully responsible for all scholarly judgements, statements, analyses, decisions, and published content. Artificial-intelligence systems cannot assume legal, ethical, or scholarly accountability.

1. Scope of the Policy

This policy applies to artificial intelligence and AI-assisted technologies used in connection with:

Manuscript drafting and revision;
Language editing and translation;
Literature searching and synthesis;
Reference generation or formatting;
Research design and protocol development;
Data collection, cleaning, coding, annotation, analysis, and interpretation;
Statistical modelling and machine learning;
Image, audio, video, signal, and multimodal analysis;
Generation or modification of text, figures, tables, images, code, prompts, or supplementary materials;
Clinical decision support, diagnosis, prognosis, risk prediction, treatment recommendation, and patient monitoring;
Peer review, editorial assessment, copyediting, production, and publication;
Post-publication correction, investigation, and research-integrity assessment.

Relevant technologies include, but are not limited to:

Generative artificial intelligence;
Large language models;
Chatbots and conversational systems;
Machine-learning and deep-learning models;
Foundation models and multimodal models;
Natural-language-processing systems;
Computer-vision and medical-imaging systems;
Clinical decision-support systems;
Automated coding, classification, prediction, and recommendation tools;
AI-assisted writing, translation, image, audio, video, or software tools;
Autonomous or semi-autonomous software agents.

2. Definitions

Artificial Intelligence

Artificial intelligence refers to computational systems designed to perform tasks commonly associated with human cognition, including language generation, prediction, classification, pattern recognition, reasoning, decision support, perception, planning, and content generation.

Generative Artificial Intelligence

Generative artificial intelligence refers to systems capable of producing new text, images, audio, video, software code, structured data, synthetic records, or other content in response to prompts, instructions, examples, or supplied data.

AI-Assisted Technology

AI-assisted technology refers to software that uses artificial intelligence to support, modify, automate, recommend, or improve a human task without necessarily generating an entire scholarly output independently.

Material Use

Material use means use that contributes substantively to the wording, content, analysis, interpretation, design, findings, images, code, or conclusions of a manuscript or research project.

Administrative or Minor Use

Administrative or minor use includes routine spelling correction, basic grammar checking, formatting, reference-library organisation, or other limited functions that do not generate substantive scholarly content or alter scientific meaning.

3. Human Accountability

Human authors remain responsible for the entire submitted and published work, including content generated, transformed, recommended, or analysed using artificial intelligence.

Authors must:

Review and verify all AI-assisted output;
Correct inaccurate, fabricated, misleading, biased, incomplete, or unsupported content;
Verify every factual statement, quotation, citation, reference, numerical result, code output, and interpretation;
Ensure that the manuscript complies with research-ethics, privacy, copyright, authorship, and data-protection requirements;
Accept responsibility for all errors or omissions, regardless of whether they originated from an AI system;
Retain sufficient human oversight over research, analysis, interpretation, and clinical judgement.

Use of an AI system does not reduce, transfer, or replace the responsibility of authors, reviewers, editors, institutions, or research sponsors.

4. Artificial Intelligence Cannot Be an Author

An artificial-intelligence system, chatbot, large language model, software agent, or other automated tool must not be listed as:

An author;
A co-author;
A corresponding author;
A contributor capable of assuming scholarly responsibility;
An institutional affiliation;
A person responsible for data, analysis, or publication decisions.

AI systems cannot meet the journal’s authorship requirements because they cannot:

Approve the final manuscript;
Assume accountability for the integrity of the work;
Respond independently to ethical or legal concerns;
Disclose conflicts of interest;
Consent to publication;
Own or transfer copyright in the same manner as a legally responsible human author.

Where an AI system has been used substantively, the use must be disclosed in the appropriate declaration and, where methodologically relevant, described in the Methods section.

5. Permitted Uses in Manuscript Preparation

Authors may use AI-assisted technologies responsibly for limited or disclosed purposes, including:

Spelling and grammar correction;
Improving linguistic clarity without changing scientific meaning;
Translation, provided that the authors verify the complete translated text;
Formatting or restructuring author-created content;
Assisting with code development or debugging;
Supporting exploratory literature searching;
Producing an initial summary that is independently checked against original sources;
Generating draft tables, figures, or documentation that are subsequently verified and revised by the authors;
Supporting data analysis when the procedures are methodologically appropriate and transparently reported.

Permitted use is conditional upon:

Human verification of all outputs;
Accurate disclosure where the use is material;
Compliance with confidentiality and data-protection requirements;
Compliance with copyright, licensing, and intellectual-property obligations;
Retention of appropriate records, prompts, versions, code, and validation materials where relevant.

Routine spelling correction and basic grammar checking that do not generate substantive content do not ordinarily require a separate declaration. Material content generation, translation, coding, analysis, image preparation, or methodological assistance must be disclosed.

6. Required AI-Use Declaration

Material use of generative AI or AI-assisted technologies must be disclosed at submission.

The declaration should identify:

The name of the tool;
The provider or developer;
The model and version, where available;
The date or period of use;
The purpose for which the tool was used;
The sections, data, analyses, images, code, or stages of work affected;
Whether prompts, system instructions, plugins, external databases, or retrieval systems were used;
The procedures used by the authors to review, verify, and revise the output;
Any known limitations or restrictions affecting reproducibility.

Suggested declaration where AI was used:

During the preparation of this manuscript, the authors used [tool name, provider, model, and version] for [specific purpose]. The tool was used in [identify the relevant sections or stages]. All AI-assisted outputs were reviewed, verified, and revised by the authors. The authors accept full responsibility for the accuracy, originality, integrity, and final content of the manuscript.

Suggested declaration where no material AI was used:

The authors declare that no generative artificial intelligence or AI-assisted technology was used to generate substantive scholarly content, analyse research data, produce scientific images, or make intellectual contributions to this manuscript.

Disclosure of appropriate AI use does not constitute misconduct and does not in itself disadvantage a manuscript. Failure to disclose material use may constitute a transparency or research-integrity concern.

7. Location of the Disclosure

AI use should be reported according to its function:

Title page or declarations section: general disclosure of material AI-assisted manuscript preparation;
Methods section: use of AI in research design, data collection, analysis, coding, modelling, diagnosis, prediction, evaluation, or clinical implementation;
Figure legend: AI-generated or materially AI-modified images, illustrations, audio, video, or visualisations;
Data and code availability statement: availability of prompts, code, model documentation, evaluation data, or model outputs;
References: citation of an AI system where it is itself a research object, software resource, model, or substantive methodological component.

Citing an AI tool does not replace the requirement to explain how it was used.

8. Prohibited Uses

Authors must not use artificial intelligence to:

Fabricate or falsify data, participants, interviews, quotations, observations, clinical records, outcomes, or results;
Create non-existent or unverifiable references;
Generate false ethics approval, trial registration, funding, authorship, or institutional information;
Manipulate images, scans, signals, figures, or data deceptively;
Conceal plagiarism, redundant publication, or inappropriate text recycling;
Present synthetic data as authentic patient or research data without disclosure;
Misrepresent automatically generated output as independently validated evidence;
Replace required clinical, ethical, legal, statistical, or methodological judgement;
Generate reviewer identities or manipulate peer review;
Create false consent forms, patient statements, survey responses, or qualitative data;
Upload confidential or identifiable information to unauthorised external systems;
Evade journal screening, research-integrity checks, or authorship accountability.

AI-assisted fabrication, falsification, plagiarism, citation manipulation, image manipulation, peer-review manipulation, or concealment of research misconduct will be treated according to the journal’s Publication Ethics policy.

9. Reference and Citation Integrity

Authors must verify every citation and reference against the original source.

AI-assisted reference generation creates a particular risk of:

Fabricated articles, books, reports, authors, journals, or DOIs;
Incorrect titles, publication years, volume numbers, or page ranges;
Misattributed quotations;
Citations that do not support the associated claim;
Misrepresentation of preprints as peer-reviewed research;
Failure to identify corrected or retracted sources.

Authors must:

Open and verify every DOI and URL;
Confirm that the source exists;
Confirm that the cited source supports the statement for which it is cited;
Check whether the source has been corrected, withdrawn, or retracted;
Format references according to the PSG Author–Date Citation Format;
Remove fabricated, duplicate, irrelevant, or unverifiable references.

Use of an AI system, reference manager, or citation generator does not relieve authors of responsibility for reference accuracy.

10. Confidentiality and Protected Information

Authors must not upload confidential, restricted, proprietary, or identifiable information to an external AI system unless:

The authors have lawful authority to do so;
Appropriate participant consent and ethics approval permit the use;
Applicable privacy and data-protection requirements are satisfied;
The tool’s data-retention, security, access, training, and deletion terms are acceptable;
Institutional or contractual approval has been obtained where required;
Reasonable safeguards prevent unauthorised retention, reuse, disclosure, or re-identification.

Restricted information includes:

Identifiable or potentially re-identifiable patient data;
Protected health information;
Unpublished manuscripts;
Confidential peer-review materials;
Unreleased datasets;
Proprietary software or source code;
Commercially sensitive information;
Institutional records;
Research-participant communications;
Information subject to legal, contractual, or security restrictions.

De-identification must be appropriate to the nature and sensitivity of the information. Removal of direct identifiers does not necessarily eliminate re-identification risk.

11. Copyright and Intellectual Property

Authors are responsible for ensuring that AI-assisted content does not infringe copyright, database rights, software licences, patents, trademarks, confidentiality obligations, or other intellectual-property rights.

Authors must not assume that AI-generated content is:

Original;
Free from copyrighted or proprietary source material;
Eligible for exclusive ownership;
Accurately attributed;
Compatible with the journal’s CC BY 4.0 licence.

Where an AI system has reproduced, transformed, or closely imitated protected material, authors must obtain any necessary permission and provide appropriate attribution.

Authors must disclose restrictions affecting:

Training datasets;
Proprietary models;
Commercial APIs;
Source code;
Model weights;
AI-generated images or media;
Third-party software components.

12. AI-Generated and AI-Modified Images

Generative AI must not be used to create or modify clinical, diagnostic, experimental, pathological, radiological, microscopic, laboratory, forensic, or other evidentiary images in a manner that misrepresents the underlying evidence.

AI-generated or materially AI-modified images may be considered only where:

The generation or modification is an explicit part of the research method;
The procedure is scientifically justified;
The tool, model, version, prompts, parameters, and processing steps are reported;
The image is labelled clearly as generated, synthetic, reconstructed, segmented, enhanced, or modified;
The original source material is retained and available for editorial assessment where applicable;
The output does not expose or fabricate identifiable patient information;
The authors possess the necessary rights and permissions.

Permissible scientific uses may include:

Image segmentation;
Noise reduction;
Reconstruction;
Registration;
Feature extraction;
Model-generated synthetic data used for validation or methodological research;
Clearly labelled conceptual or explanatory illustrations.

All processing must be described sufficiently to distinguish original evidence from computationally generated or modified content.

13. Synthetic Data

Synthetic data must not be presented as real patient, participant, clinical, observational, or experimental data.

Research involving synthetic data must report:

The purpose for generating synthetic data;
The model or procedure used;
The source data used to train or condition the generator;
The extent to which source data included protected or identifiable information;
Privacy and re-identification risk assessment;
Methods used to evaluate fidelity, diversity, utility, bias, and disclosure risk;
Whether synthetic data were used for training, testing, augmentation, simulation, or demonstration;
The limitations of using synthetic rather than real-world data.

Authors must not claim that synthetic data are automatically anonymous, unbiased, representative, or privacy-preserving without appropriate evidence.

14. Research Involving AI and Machine Learning

Studies involving artificial intelligence or machine learning must provide sufficient information for critical assessment, reproducibility, and evaluation of clinical or public-health relevance.

Authors should report, where applicable:

The intended use of the system;
The clinical, public-health, administrative, or research problem addressed;
The intended users and affected population;
The intended decision context and workflow;
Whether the system supports, recommends, or makes decisions;
The level of human oversight;
The source and characteristics of all datasets;
The dates and geographic origins of data collection;
Inclusion and exclusion criteria;
Data labelling, annotation, and reference standards;
Data cleaning, preprocessing, transformation, and missing-data handling;
The model architecture and software environment;
Training procedures and hyperparameters;
Training, validation, and test-set separation;
Measures used to prevent data leakage;
Comparator systems or standards of care;
Internal, external, temporal, geographic, or multisite validation;
Performance measures and uncertainty estimates;
Calibration, discrimination, clinical utility, and error analysis;
Subgroup performance and fairness assessment;
Model limitations and failure cases;
Availability of code, models, weights, prompts, documentation, and data.

Performance measured only on training data does not establish clinical validity. Claims of safety, superiority, effectiveness, generalisability, or implementation readiness must be supported by the study design and evidence.

15. Data Separation and Leakage

Authors must explain how training, tuning, validation, and test datasets were created and separated.

The manuscript must address potential leakage arising from:

The same patient appearing in more than one dataset;
Multiple samples, images, encounters, or time points from the same individual;
Institutional overlap;
Temporal overlap;
Preprocessing performed before dataset separation;
Feature selection using test data;
Repeated access to the test set during model development;
Public benchmark contamination;
Training-data exposure in a proprietary foundation model;
Prompt or answer contamination in large language model evaluation.

Undisclosed leakage that materially inflates performance may constitute unreliable reporting or research misconduct.

16. Validation and Generalisability

Authors must distinguish clearly among:

Model development;
Internal validation;
External validation;
Temporal validation;
Geographic validation;
Prospective clinical evaluation;
Real-world implementation;
Post-deployment monitoring.

Authors must not describe internal testing as independent external validation.

Where external validation has not been performed, the manuscript must avoid unsupported claims of broad clinical generalisability.

Validation datasets should be sufficiently independent of development data and should reflect the population, prevalence, institutions, devices, workflows, and conditions relevant to the proposed use.

17. Bias, Fairness, and Equity

Authors must consider whether AI systems may produce unequal performance, access, benefit, or harm across relevant populations.

Where data permit, authors should evaluate performance according to characteristics such as:

Age;
Sex and gender;
Race and ethnicity;
Language;
Disability;
Geography;
Socioeconomic status;
Healthcare setting;
Disease severity;
Device type or data source;
Other clinically or socially relevant groups.

Authors should report:

Representation in the development and evaluation data;
Subgroup sample sizes;
Subgroup performance and uncertainty;
Potential sources of historical, measurement, selection, annotation, or deployment bias;
Methods used to identify or mitigate unfairness;
Remaining limitations and potential consequences for health equity.

Claims that a model is fair, unbiased, neutral, or universally applicable require empirical support and must be defined using appropriate criteria.

18. Explainability, Transparency, and Human Oversight

Authors should explain how users understand, interpret, challenge, or override the outputs of an AI system.

Where relevant, manuscripts should describe:

The explanation or interpretability method used;
The intended audience for the explanation;
Whether the explanation reflects the actual model process or provides only an approximation;
Evidence that the explanation is meaningful to clinicians, patients, or other intended users;
The role of human judgement;
Override procedures;
Escalation and error-reporting mechanisms;
Responsibility for final clinical or operational decisions.

Explainability must not be presented as proof of model accuracy, causal validity, fairness, or safety.

19. Clinical Safety and Intended Use

Studies involving clinical AI must define the system’s intended use and the consequences of incorrect, delayed, or missing outputs.

Authors should assess, where relevant:

False-positive and false-negative consequences;
Unsafe recommendations;
Automation bias;
Over-reliance by clinicians or patients;
Delayed care or inappropriate escalation;
Workflow disruption;
Alert fatigue;
Cybersecurity and system failure;
Changes in model performance over time;
Requirements for clinical monitoring and human intervention.

Research findings must not be presented as regulatory approval, clinical certification, or evidence of deployment readiness unless the relevant requirements have been satisfied.

20. Generative AI and Large Language Model Studies

Studies involving generative AI, large language models, foundation models, or conversational agents should report:

The developer or provider;
The model name;
The model version or release;
The date of access and evaluation;
The access method, interface, API, or deployment environment;
System instructions and prompt templates;
Prompt-development and prompt-selection procedures;
Sampling parameters, temperature, token limits, or other available settings;
The number of repeated generations;
Whether conversation history or memory was enabled;
The use of retrieval-augmented generation;
External databases, tools, plugins, agents, or search systems;
The procedure for selecting or excluding outputs;
Human and automated evaluation methods;
Evaluator qualifications and blinding;
Inter-rater reliability where applicable;
Methods used to evaluate factual accuracy, hallucination, bias, safety, and clinical risk;
Measures used to protect confidential or health-related information.

Because proprietary models may change without notice, authors should archive prompts, outputs, timestamps, screenshots, API records, or other documentation sufficient to support verification where legally and technically permissible.

Single-query demonstrations should not be presented as stable evidence of model capability without appropriate justification and repeated evaluation.

21. Reproducibility and Record Retention

Authors should retain sufficient documentation to permit evaluation and, where possible, reproduction of AI-related research.

Relevant records may include:

Study protocols;
Data dictionaries;
Preprocessing scripts;
Annotation guidelines;
Training and validation code;
Model configurations;
Hyperparameters;
Random seeds;
Software and library versions;
Model weights;
Prompts and system instructions;
Generated outputs;
Evaluation rubrics;
Audit trails;
Version histories;
Deployment and monitoring records.

Where materials cannot be shared because of legal, ethical, security, privacy, licensing, or commercial restrictions, authors must explain the restriction clearly and identify what information can be made available.

22. Data, Code, Model, and Prompt Availability

AI-related research articles must include an appropriate availability statement.

The statement should identify whether the following are available:

Training, validation, and test data;
Data dictionaries and metadata;
Preprocessing and analysis code;
Model architecture and configuration;
Model weights or checkpoints;
Software dependencies;
Prompts and system instructions;
Evaluation datasets and scoring criteria;
Protocols and statistical analysis plans;
Model cards, data sheets, or technical documentation.

A statement such as “data available on request” should explain the conditions, responsible party, review process, and legal or ethical limitations applicable to access.

23. Reporting Guidelines

Authors must follow the reporting guideline appropriate to the research design and upload the relevant checklist where required.

AI-related guidelines may include:

CONSORT-AI for clinical trials involving an AI intervention;
SPIRIT-AI for protocols of clinical trials involving AI;
TRIPOD+AI for studies developing or evaluating clinical prediction models using regression or machine-learning methods;
TRIPOD-LLM for studies using large language models where applicable;
STARD-AI for diagnostic-accuracy studies using AI;
DECIDE-AI for early-stage clinical evaluation of AI decision-support systems;
CLAIM for artificial intelligence in medical imaging;
Other relevant EQUATOR Network guidance.

Use of a reporting checklist does not correct a weak study design. Authors remain responsible for methodological validity and complete reporting.

24. Ethical Approval for AI Research

Research involving identifiable human data, health records, biological information, patient interactions, clinical decision-making, or human participants requires appropriate ethics review unless a valid exemption applies.

Authors must report:

The approving ethics committee or institutional review board;
The approval number;
The informed-consent procedure;
Any waiver of consent or ethics review;
The legal and ethical basis for data access and reuse;
Privacy and security safeguards;
Whether commercial or external AI providers received participant data;
Whether data were used to train or improve external models.

Prior ethics approval for the original collection of data does not automatically authorise every subsequent AI use, data linkage, commercial transfer, model training, or international data transfer.

25. AI Used in Clinical Care or Research Interventions

Where an AI system affects clinical decisions, participant treatment, risk classification, recruitment, monitoring, or intervention allocation, authors must explain:

Whether the system was investigational or approved for the intended use;
The qualifications of individuals supervising its use;
How outputs were reviewed and acted upon;
Whether participants were informed of AI involvement;
How errors, adverse events, and unsafe outputs were managed;
Whether a clinician could override the system;
How responsibility for decisions was assigned;
What monitoring and stopping procedures were in place.

AI systems must not be used in a manner that exposes participants or patients to unjustified risk.

26. Use of AI by Reviewers

Submitted manuscripts and associated files are confidential scholarly communications.

Reviewers must not upload any part of a manuscript, supplementary file, dataset, source code, author response, or reviewer report to a public or third-party AI system where confidentiality, intellectual-property protection, security, and deletion cannot be assured.

An AI system cannot act as a reviewer and cannot replace expert scholarly judgement.

Any proposed use of AI during peer review requires prior written authorisation from the editorial office. Where authorised, the reviewer must:

Identify the tool and version;
Explain the purpose of use;
Protect all confidential information;
Verify the output independently;
Disclose the use to the journal;
Accept full responsibility for the review.

Reviewers must not use AI to infer author identities, fabricate criticisms, generate unsupported recommendations, or evaluate confidential patient information without authorisation.

27. Use of AI by Editors

Editors must protect the confidentiality and integrity of submitted manuscripts.

Editors must not upload manuscripts, reviews, author identities, reviewer identities, supplementary files, or confidential correspondence to an external AI system where confidentiality cannot be assured, except with appropriate authorisation and safeguards.

AI may support limited administrative or technical functions, but it must not make autonomous decisions concerning:

Editorial acceptance or rejection;
Reviewer selection;
Research misconduct;
Ethical acceptability;
Clinical validity;
Authorship or conflict-of-interest disputes;
Corrections or retractions.

The responsible editor must review and approve every editorial decision.

28. Use of AI by the Journal and Publisher

The journal or publisher may use automated or AI-assisted systems for limited operational purposes, including:

Similarity screening;
Reference checking;
Metadata preparation;
Language or formatting assistance;
Image-integrity screening;
Technical compliance checks;
Accessibility conversion;
Production-quality assurance.

Such tools are used to support, not replace, human editorial and production oversight.

The journal will not base a rejection, allegation of misconduct, correction, or retraction solely on the output of an automated AI-detection tool. Relevant evidence must be assessed by qualified human editors or experts.

Where AI is used materially in copyediting, production, summaries, educational content, promotional materials, or other journal-generated content, the journal remains responsible for verifying accuracy and identifying AI use where appropriate.

29. Automated AI-Use Detection

Automated systems that attempt to identify AI-generated text may produce false-positive and false-negative results.

The journal may use such systems as preliminary screening tools but will not treat an automated score as conclusive evidence of:

Undisclosed AI use;
Plagiarism;
Fabrication;
Authorship misconduct;
Research misconduct.

Where concerns arise, the journal may request clarification, version history, prompts, drafts, source files, data, code, or other relevant documentation before reaching a decision.

30. Undisclosed or Inappropriate AI Use

Possible concerns include:

Material AI use not disclosed by the authors;
AI-generated references that do not exist;
Fabricated text, data, participants, quotations, or results;
AI-generated or modified images presented as authentic evidence;
Use of confidential patient information without authorisation;
Misrepresentation of synthetic data;
Misleading claims regarding model performance or validation;
Failure to report model version, prompts, data sources, or evaluation methods;
Use of AI to manipulate authorship or peer review;
Use of copyrighted or proprietary content without permission.

Minor or inadvertent disclosure deficiencies may be corrected where they do not affect the integrity of the research. Serious or deliberate misconduct may result in rejection, withdrawal of acceptance, correction, expression of concern, retraction, institutional notification, or other proportionate action.

31. Editorial Investigation

Where an AI-related concern is identified, the journal may:

Conduct a preliminary confidential assessment;
Request an explanation from the corresponding author;
Request prompts, outputs, drafts, version histories, data, code, model documentation, or original images;
Seek independent technical, statistical, ethical, legal, or clinical advice;
Pause peer review, acceptance, production, or publication;
Contact co-authors, institutions, ethics committees, funders, registries, or relevant authorities;
Issue a correction, expression of concern, retraction, or removal notice where necessary.

The journal will assess concerns fairly, confidentially, and proportionately. An allegation will not be treated as established misconduct solely because an AI-detection system or anonymous complainant raises a concern.

32. Post-Publication Responsibilities

Authors must notify the journal promptly if they discover that AI-assisted content has introduced:

A factual error;
A fabricated or incorrect reference;
An analytical or coding error;
A privacy or confidentiality breach;
A misleading image or figure;
A material error in model reporting;
An incorrect conclusion;
Another issue affecting the reliability or integrity of the article.

The journal may publish an appropriate correction or other post-publication notice. Where the findings are no longer reliable, retraction may be necessary regardless of whether the problem resulted from intentional misconduct or unintentional error.

33. Sanctions and Corrective Actions

Where inappropriate AI use or related research-integrity misconduct is substantiated, the journal may take one or more actions:

Request clarification or disclosure;
Require correction of the manuscript;
Return the submission for technical revision;
Reject the manuscript;
Withdraw an acceptance decision;
Publish a correction or expression of concern;
Retract the article;
Restrict future submissions for a defined period;
Remove an individual from reviewer or editorial roles;
Notify institutions, ethics committees, funders, registries, or other journals;
Refer a serious legal, privacy, safety, or security matter to an appropriate authority.

Actions will take account of the seriousness of the issue, intent, effect on the scholarly record, risk to patients or the public, author cooperation, recurrence, and available evidence.

34. Appeals

Authors may appeal an AI-related editorial decision where they believe that a material procedural, factual, methodological, or technical error occurred.

An appeal should:

Identify the manuscript and decision;
Explain the specific basis of the appeal;
Provide relevant evidence, documentation, prompts, code, data, or version history;
Address the journal’s concerns professionally and directly.

General disagreement with the policy or denial of AI use without supporting evidence does not by itself establish grounds for appeal.

35. Recommended External Standards

The journal’s policy is informed by recognised guidance concerning artificial intelligence, medical publishing, reporting quality, research integrity, and publication ethics.

36. Policy Review

Artificial-intelligence technologies and their associated ethical, clinical, legal, and methodological risks evolve rapidly. The journal may revise this policy to reflect developments in:

Medical and clinical artificial intelligence;
Generative AI and autonomous agents;
Research-integrity standards;
Privacy and data-protection law;
Medical-device and clinical regulation;
Copyright and intellectual-property law;
Peer-review and editorial practice;
Recognised reporting guidelines.

Authors, reviewers, and editors are responsible for consulting the current version of this policy and complying with any additional institutional, legal, ethical, contractual, or regulatory requirements applicable to their work.