1. Purpose and scope

Artificial intelligence tools are increasingly prevalent in scientific writing and manuscript review. They can be helpful, but they can also fabricate references, reproduce existing texts, or expose confidential data. This policy specifies what the Cameroonian Annals of Laboratory Medicine, referred to here as "the journal," permits, prohibits, and requires to be disclosed.

It is addressed to authors, reviewers, editors, members of the editorial secretariat, and anyone involved in manuscript processing.

By artificial intelligence tools, we mean generative AI systems—systems that produce text, images, tables, or code based on instructions, particularly conversational assistants based on large language models. Translation and paraphrasing tools based on these models are included. Simple spell-checkers and bibliographic management software are not covered.

Scientific scope: clinical biology, medical biochemistry, hematology, immunology, medical microbiology, molecular biology, toxicology.

2. Principles

• Human accountability. A person is always responsible for what they sign, whether an article, a review report, or a decision.
• Transparency. Any use that has influenced the content of a document is disclosed.
• Confidentiality. A manuscript under review is never entrusted to a tool whose data use cannot be controlled.
• Integrity. No tool replaces the observation, analysis, or critical thinking of researchers.
• Benevolence. The journal does not reject these tools on principle. It recognizes that they can help researchers who do not write in their native language, and honest disclosure is never grounds for rejection in itself.

This policy is inspired by the recommendations of the International Committee of Medical Journal Editors, the position of the Committee on Publication Ethics on authorship criteria and artificial intelligence tools, and the guidance of the World Association of Medical Editors. It will be adapted as these references evolve.

3. Authors

3.1 A tool is not an author

An artificial intelligence tool cannot be listed among the authors. It cannot approve the final version, be responsible for the integrity of the work, or declare conflicts of interest—requirements of authorship criteria.

3.2 Permitted uses, to be disclosed

Authors may use these tools to improve the language of their text, translate it, rephrase it, or refine its clarity. They may also use them to understand the conditions for applying a statistical test before choosing the appropriate one, verifying their choice afterward themselves or with a statistician. They carefully review and correct everything the tool proposes, as they remain responsible for every sentence. Such use is mentioned in the section provided for this purpose, as indicated in section 3.7.

3.3 Prohibited uses

• Producing, completing, or modifying data or results.
• Writing the interpretation or conclusions in place of the authors.
• Generating references. Authors cite only sources they have read and verified to exist and be accurate, as these tools sometimes produce plausible but fabricated references.
• Delegating to the tool the production of text or content. This would amount to attributing part of the authorship to the tool, which ICMJE recommendations exclude.
• Entering into a public tool data that could identify a patient, or information covered by an ethics committee opinion.

Before using a tool, authors verify its terms of use and settings. Data entered should serve only the current request and not be reused to train the model. This applies also to their own unpublished manuscripts.

3.4 Images, diagrams, and data

Images representing research data. No generative AI tool can create or modify an image intended to show an observation or result the study did not obtain as presented: blood smears, cultures, electrophoresis, cytograms, imaging, sections, clinical photographs. The same rule applies to data. This is absolute, except when image generation or modification is the study's own object and is described in the Methods section. Only adjustments of brightness, contrast, or color applied to the entire image, which do not mask or add information, are permitted.

Diagrams, decision trees, timelines, and infographics. For figures not representing research data, a tool may be used provided that:

• the figure provides useful information that does not repeat the text or a table and relates precisely to the article's subject. Generic or purely decorative illustrations are excluded;
• tool use is disclosed in the caption, with its name and version;
• authors can guarantee the accuracy, originality, and free availability of rights to the figure, as a tool may unknowingly reproduce protected elements;
• successive versions of the figure and instructions given to the tool are kept and provided to the editorial office upon request.

Any image or video generated by artificial intelligence is clearly designated as such. No generated image depicts a real person, identifiable brand or product, or reproduces a protected image. No synthetic voice reproduces the voice of an existing person. The number of figures remains proportionate to their usefulness. The editorial office may reduce, modify, or reject images that do not meet these criteria.

3.5 Artificial intelligence as a research method

When artificial intelligence is part of the method—for example, for automated smear reading, cytology image classification, or construction of a prediction model—it is described in the Methods section. Authors specify the tool or algorithm, its version, training and validation data, performance achieved, and how results were verified by professionals. They follow the appropriate reporting guideline, such as CLAIM for imaging or TRIPOD+AI for prediction models, in addition to the guideline corresponding to the study type. This use does not substitute for the disclosure in section 3.7: it supplements it if tools were also used for writing.

3.6 Originality and plagiarism

Text produced by a tool may reproduce, without stating so, passages or ideas from existing works. Authors ensure their manuscript is original and properly referenced. Similarity checks apply the same way to all manuscripts. A fabricated reference or passage reproduced without citation constitutes a breach of integrity, regardless of its origin.

3.7 How to disclose

Authors fill in the "Artificial intelligence use" section placed at the end of the manuscript, after data availability. They indicate:

• the tool's name, version, and provider;
• the period of use;
• the purpose of use and the parts of text concerned;
• how they supervised, verified, and corrected the result.

If no tool was used, the section states this. Sample formulations are provided in Appendix B.

4. Reviewers

4.1 An evaluation conducted by the reviewer

A reviewer is invited for their expertise and judgment. The journal therefore expects them to read the manuscript, evaluate it themselves, and write their report themselves. Scientific evaluation is not delegated to a tool.

4.2 Full responsibility

Everything the reviewer submits to the journal is their sole responsibility: content, accuracy, tone, and recommendations. A report containing errors or fabricated references is the responsibility of the person who signed it, even if a tool was involved. A report that appears to have been produced primarily by a tool will not be accepted. The associate editor may ask the reviewer to redo it and may remove them from the reviewer pool.

4.3 Confidentiality

It is prohibited to upload into an artificial intelligence tool the manuscript, any part of it, its figures, abstract, or data, or the reports of other reviewers. This applies even when the tool claims to be private or states it does not retain content. It protects authors, patients whose data may appear in manuscripts, and copyright in unpublished text.

Artificial intelligence functions integrated into a PDF reader, word processor, or email client, such as automatic summarization, are not activated on the manuscript.

4.4 Limited linguistic assistance

A reviewer whose working language is not French or English may use a tool to correct or translate their own comments, provided that three conditions are met. They enter only their remarks into it, without any manuscript passage, data, or identifying information about the authors. They review and validate the result. They inform the associate editor of this use in the confidential section of their report.

5. Editors and editorial secretariat

• The confidentiality rules of section 4.3 apply to editors and the secretariat. No manuscript or editorial correspondence is transmitted to an artificial intelligence tool.
• Editorial decisions are never delegated to a tool. Decision letters and exchanges with authors are written by a person.
• Control tools—for example, for text similarity, image analysis, or reference verification—may assist the secretariat. Their results serve as evidence, and a person examines them before any action. These tools respect manuscript confidentiality.
• The journal may use software to detect AI-generated text. These produce false positives and false negatives, and they often penalize those writing in a second language. Their result is evidence that a person examines, and it is never sufficient alone to question an author or reviewer.
• The journal may use closed tools to produce missing metadata, such as translating an abstract or keywords. The manuscript deposited there is not used to train the model or produce other content, and a person verifies the result. Any other use in the production chain is indicated.

6. Breaches

Breaches include: undisclosed use that influenced content, fabricated references or data, images generated or edited without disclosure, and uploading a confidential manuscript into a tool. Processing follows the procedures of the Committee on Publication Ethics described in the journal's ethics charter. The response is proportionate. It ranges from requesting explanation to correcting or rejecting a manuscript under review, to issuing an erratum or retraction after publication. For a reviewer, it may extend to removal from the reviewer pool and, in serious cases, notifying their institution.

When use contrary to this policy is established or reasonably suspected, the journal may reject the manuscript. Failure to disclose may, depending on circumstances, be treated as misconduct. An editorial committee member with reasonable and documented suspicion against an author or reviewer informs the editor-in-chief.

A person who discovers an error related to a tool themselves and reports it promptly will be treated with benevolence.

7. Evolution of the policy

This field evolves rapidly. The policy is reviewed at least once a year, and sooner if a major international recommendation requires it. Any question may be addressed to the editorial secretariat at contact@jscbc-calm.cm.

Appendix A. Summary of rules

Actor	Permitted	Prohibited	To disclose
Authors	Improve language, translate, rephrase, understand statistical test conditions Diagrams unrelated to data, with disclosure in caption Use AI as research method, if described	List a tool as an author Generate data, results, references, or data images Enter patient data into a public tool	All use influencing content, in the dedicated section AI used as method, in Methods
Reviewers	Evaluate the manuscript themselves Correct language of their own comments, without manuscript content	Upload the manuscript or part into a tool Activate automatic summary on the manuscript Submit a report written primarily by a tool	Any linguistic assistance, to the associate editor
Editors and secretariat	Use control tools, with human verification	Upload a manuscript into a tool Delegate decision to a tool Rely on AI-generated text detection	All tools used in the production chain

Appendix B. Sample formulations

Situation	English
Authors, editorial use	"Tool name, version, developer" was used in "month, year" to improve the language of the text. The authors reviewed and edited all content and take full responsibility for it.
Authors, no use	No artificial intelligence tools were used in the preparation of this manuscript.
Authors, generated diagram	Diagram created with "tool, version". The authors guarantee the accuracy of the content and the right to use it.
Authors, AI as method	An artificial intelligence algorithm is part of the study method. It is described in the Methods section.
Reviewer, linguistic assistance	I used "tool, version" only to correct the language of my own comments. No content from the manuscript was entered into it. The scientific assessment is entirely my own.