Our AI Use Policy

We use some artificial-intelligence tools in our work, and readers, clients, and colleagues are entitled to know where and how. Our guiding principle is simple:

AI assists our work; it does not replace human judgement. A designated human remains accountable for every published, shared, or client-facing result.

How we may use AI

Data analysis and research

Our primary use of AI is to support the checking, processing, and analysis of datasets. This work is carried out by a trained and qualified data analyst, or by a team of data-science professionals.

These professionals use designated workflows and skillsets in tools such as Claude, ChatGPT, and, where a more flexible or exploratory approach is appropriate, Gemini. Each task is scoped, assigned, and reviewed by a responsible human before the work begins.

AI may assist with:

  • Data cleaning, sanitisation, organisation, and quality checks.
  • Statistical and inferential analysis.
  • Cross-comparisons with published datasets and other reliable sources.
  • R, Python, and other coding required to process or analyse data.
  • Reproducible calculations, visualisations, and technical documentation.
  • Identifying inconsistencies, gaps, or possible errors for human investigation.

AI-generated code, calculations, and interpretations are treated as provisional. A qualified person must review the methods, inputs, outputs, assumptions, and limitations before conclusions are accepted.

The resulting conclusions and summaries are discussed with the news team so that the people responsible for the work understand what the analysis shows, what it does not show, and how confidently it can be reported.

Newsroom and research support

AI may be used to assist with transcription, document search, background research, language processing, and the organisation of information.

We may also run newsroom automation tasks to support research into language, content, books, newsfeeds, and related topics. These outputs are inspected and reviewed by an actual human before they inform published work or editorial decisions.

Narration and translation

AI may be used to:

  • Produce an audio narration of an article, provided it is clearly labelled as a synthetic voice.
  • Draft translations between English and Dhivehi, provided a person reviews the translation before publication.
  • Support other language or accessibility features, provided the result is checked for accuracy and suitability.

Summaries and client-facing services

We may provide short, clearly labelled AI-generated summaries alongside human-written summaries. An AI summary will never replace the original article or the human-written editorial summary.

We may also use AI in client-facing summaries and server-side automations. These systems must have a designated human owner, appropriate monitoring, and a process for correcting inaccurate or misleading outputs. Where appropriate, users and clients will be told when an output has been generated or materially assisted by AI.

Where we do not use AI

We do not use AI to:

  • Publish AI-written articles as though they were reported and written by a journalist.
  • Invent or fabricate quotes, sources, facts, evidence, or documentary material.
  • Make final decisions about publication.
  • Override an editor, journalist, analyst, or other responsible human.
  • Determine the overall direction, editorial priorities, opinions, or journalistic purpose of an endeavour.
  • Replace the judgement and responsibility of the people doing the reporting, analysis, editing, or review.

AI may assist with the mechanics of a task, but humans decide what matters, what is credible, what should be published, and how it should be presented.

Human oversight and accountability

Every AI-assisted task must have a responsible human owner. That person is accountable for:

  • Defining the task and selecting an appropriate workflow.
  • Checking the quality and suitability of the source material.
  • Reviewing AI-generated text, code, analysis, translations, summaries, and other outputs.
  • Confirming important claims against reliable sources, source data, documentation, or reproducible checks.
  • Identifying uncertainty, limitations, bias, and possible errors.
  • Approving the final result or deciding that it must not be used.

AI output is considered unverified until a human has checked it. No output is accepted merely because it is plausible, confidently written, or technically sophisticated.

Data protection and security

We do not enter confidential, personal, unpublished, or legally restricted information into an AI tool unless the relevant workflow has been approved and appropriate safeguards are in place.

Where AI is used to process data, the responsible team must consider privacy, security, licensing, confidentiality, retention, and the terms under which the tool or service handles that data.

Disclosure and corrections

Where AI has played a meaningful role in a specific story, product, summary, or service, we disclose that use in a clear and appropriate way.

If an AI-assisted result is wrong, misleading, or incomplete, our normal editorial, client, and corrections processes apply. AI use does not reduce or transfer our responsibility.

As these tools and our workflows develop, we will review and update this policy.