AI Policy

AI tools do not hold editorial authority at JOSEPH MMWA.

Decisions about which stories to pursue, how to interpret evidence, which sources to trust, what claims to publish and whether a report is ready for publication remain the responsibility of the human editor.

AI-generated or AI-assisted material is not automatically suitable for publication. It must be assessed in context, with particular attention to factual accuracy, meaning, source attribution and the potential consequences of errors.

The extent of review should reflect the nature and sensitivity of the material. Health claims, medical research, statistics, treatment information and breaking public-health developments require particular scrutiny.

AI tools may assist with:

• Research support and the organisation of information.
• Summarising documents and scientific literature for editorial review.
• Identifying factual claims, inconsistencies and questions that require further investigation.
• Developing preliminary article drafts, headlines, outlines and explainers.
• Editing, proofreading, formatting and improving readability.
• Transcription and translation.
• Supporting editorial workflows, including the organisation of news alerts and the preparation of material for review.
• Assisting with the presentation of information and other production tasks.

These are potential or permitted applications, not a guarantee that every tool or application is used in every article. The use of AI does not, by itself, establish the accuracy, originality or completeness of the resulting material.

AI systems can produce false statements, misinterpret evidence, omit important context, generate nonexistent references or provide outdated information with apparent confidence.

For this reason, AI output is treated as material requiring assessment, not as proof that a claim is true.

Where appropriate, I prioritise primary sources, including original research papers, clinical trial records, official health advisories, regulatory documents, public-health authorities and statements from relevant institutions. Secondary reporting may also be used where it provides relevant context or original reporting, with appropriate attribution.

A source suggested by an AI system must be checked before it is relied upon. Where possible, I examine the original document rather than relying solely on an AI-generated summary.

AI-assisted fact-checking is not infallible. It can miss errors, misread documents or incorrectly assess whether a claim is supported. It therefore does not replace source verification and editorial judgment.

Where a claim cannot be adequately verified, I may qualify it, identify the uncertainty, seek further evidence or exclude it from publication.

Health journalism requires particular care because inaccurate information can influence medical decisions, public understanding and confidence in healthcare.

When AI tools assist with reporting on diseases, medicines, vaccines, clinical trials, treatments, medical devices, public-health guidance or medical risks, I assess the underlying evidence rather than relying on the tool's interpretation alone.

This includes distinguishing, where relevant, between:

• Laboratory or animal research and evidence from human studies.
• Early-stage research and established clinical evidence.
• Statistical associations and demonstrated causal relationships.
• Relative risk reductions and absolute changes in risk.
• Study findings and claims that go beyond what the research establishes.
• Regulatory approval, research results and evidence of real-world effectiveness.

I do not treat an AI-generated medical explanation as a substitute for reliable scientific evidence or qualified clinical judgment. The strength of the available evidence and the potential consequences of an error inform the level of scrutiny required before publication.

AI tools may help locate, organise or summarise information, but they are not a substitute for identifying the origins of that information.

I do not knowingly present another journalist's reporting, a researcher's findings or an institution's original work as my own. When a story relies materially on external reporting, research or institutional information, I seek to attribute the relevant source clearly.

AI-assisted rewriting, summarising or paraphrasing does not remove the responsibility to acknowledge original work. Nor does the fact that a tool produces new wording establish that the underlying reporting is original.

I do not knowingly use AI to fabricate sources, quotations, expert opinions, interviews, research findings, statistics, citations or other evidence.

If a quotation cannot be traced to a reliable source, it must not be presented as an authentic quotation merely because an AI tool generated it.

AI may help develop an initial draft or improve the clarity of existing material. However, generated text can introduce unsupported claims, distort the meaning of a source, remove important qualifications or make uncertain findings sound conclusive.

AI-assisted drafts therefore require editorial assessment before publication. Where material claims are involved, the underlying evidence must be checked.

I remain responsible for the final wording, headline, context and conclusions of published journalism. I do not regard fluency, confidence or professional-sounding language as evidence of factual accuracy.

The final article must meet the same editorial standards regardless of whether AI contributed to its preparation.

AI-generated images and synthetic media can mislead readers when presented as authentic records of real events or people.

I do not knowingly present AI-generated imagery as an authentic photograph of a real person, event, clinical procedure, disease outbreak or medical situation.

Where synthetic imagery is used and its artificial nature is material to how readers understand the content, I will seek to identify it clearly. Illustrations and reconstructions should not be used in ways that create a false impression of documentary evidence.

Photographs, graphics and other visual material obtained from external sources remain subject to appropriate attribution, permissions where required and checks for contextual accuracy.

The use of AI to create, edit or enhance an image must not be used to fabricate evidence or materially misrepresent a real event.

Readers should not be misled about how a story was produced or what evidence supports it.

Routine assistance with proofreading, formatting, transcription or similar tasks does not necessarily require a separate disclosure. However, where AI makes a substantial contribution to published material, creates synthetic media, or otherwise plays a role that could reasonably affect a reader's understanding of the content, I will consider whether clear disclosure is warranted.

Disclosure decisions should reflect the significance of the AI contribution and the risk of misleading readers. They should not be used to imply that an article is accurate simply because AI use has been disclosed.

This policy does not mean every article will carry an AI label. Nor does the absence of such a label establish that no AI tool was used during the editorial process.

Journalism can involve confidential sources, unpublished material, sensitive health information and personal data.

I recognise that entering information into an AI service may expose it to processing under that service's terms, settings and data-handling practices. I therefore seek to avoid submitting confidential source identities, sensitive personal information or unpublished material to AI tools unless there is an appropriate basis and adequate safeguards for doing so.

Where possible, sensitive information should be minimised or de-identified before using an AI tool. Any use must take account of applicable privacy obligations, source protection and the need to preserve journalistic trust.

AI tools should not be treated as confidential simply because they are used for editorial work.

AI systems can reproduce stereotypes, omit underrepresented perspectives or reflect imbalances in the material on which they were developed or trained.

I recognise that these limitations may affect how people, countries, communities, diseases and health systems are described.

AI-assisted material should be assessed for misleading generalisations, stigmatising language, unsupported assumptions and unfair representation. Particular care is necessary when reporting on vulnerable populations, infectious diseases, disability, mental health and other sensitive health issues.

AI-generated language must not be allowed to replace careful consideration of the people and communities affected by a story.

No AI tool can guarantee that a report is accurate, balanced, complete, original or free from bias.

AI can assist with tasks, but it cannot independently establish whether a source is trustworthy, whether a claim is adequately supported or whether a story serves the public interest in a particular context.

The responsibility for published journalism remains with JOSEPH MMWA. It cannot be transferred to an AI system, a software provider or a generated response.

Where an AI-assisted error is identified in published material, the same correction and accountability principles that apply to other editorial errors apply.

AI technology and its use in journalism continue to evolve. This policy may be updated as editorial practices, available tools, applicable requirements and the risks associated with AI change.

Any update should preserve the central principle that technological assistance does not reduce editorial responsibility.

AI is a tool within the editorial process, not a substitute for journalism.

It may help organise information, examine material, develop drafts and improve presentation. It cannot replace the responsibility to verify evidence, attribute original work fairly, communicate uncertainty and consider the consequences of publication.

The technology may assist the work. Responsibility for the journalism remains human.