Ai doesn't just change how journalism is made. It changes whose stories get told.

The same four principles apply in media as in any other sector — but journalism gives each one a specific character. Consent involves publishers and sources as well as audiences. Representation is the core editorial standard the industry has always claimed to hold itself to. Provenance is the difference between journalism and content. Accountability is what distinguishes a newsroom from a content farm.

There is a structural conflict in media Ai that doesn't exist in most other sectors. The commercial objective — engagement, clicks, time-on-site, shares — and the editorial objective — accuracy, public interest, democratic function — can pull in opposite directions. Most media Ai tools are optimised for the former. Few organisations have explicitly decided which one takes precedence, or who is responsible for maintaining that decision over time.

Before commissioning or deploying any Ai tool in a newsroom or media organisation — and revisited at each renewal, upgrade, or change of use — these five questions must have documented answers. Agreed at editorial leadership level. Not left to the vendor.

1. 01

   Was the training data assembled with publisher and source consent? Consent

   The vast majority of large language models were trained on journalism scraped without permission. If your editorial Ai tool was trained on news archives, the organisation needs to understand whose content was used and on what basis. This is both an ethical and a legal question — and the answer to it affects every piece of content the tool produces.

   Training data rightsPublisher consentCopyright compliance
2. 02

   Does the training corpus represent the full diversity of the communities this organisation serves? Representation

   Not just English-language sources, not just Western geopolitical perspectives, not just the communities that traditional news archives covered well. If the tool is going to help produce journalism for a diverse readership, the data it was built on needs to reflect that diversity — or the output will systematically underserve parts of the audience.

   Geographic diversitySource diversityLanguage coverage
3. 03

   Can every Ai output be traced to a verifiable source that an editor could defend in a corrections process? Provenance

   If the answer to this is no — if the tool produces summaries, headlines, captions, or content where the sourcing is opaque — then the organisation is publishing content it cannot stand behind. That is an editorial standards failure regardless of whether a machine or a journalist produced it.

   Source traceabilityHallucination testingCorrections policy
4. 04

   Is there a named editor accountable for every category of Ai-influenced editorial decision? Accountability

   Not a policy document. A person. If an Ai tool selects which stories are promoted, which headlines run, or which content is surfaced to which audience segment — and no named editor is responsible for that decision — then editorial governance has been quietly handed to an algorithm. That is a decision that should be made explicitly, not by default.

   Editorial governanceNamed accountabilityOversight structure
5. 05

   Has the organisation explicitly decided whether its Ai systems are optimised for engagement or for editorial quality — and does that decision align with its stated values? Accountability

   This is the question most media organisations haven't asked out loud. The answer shapes every downstream ethical consequence of Ai deployment. A news organisation that says it is committed to public-interest journalism, and then deploys a recommendation system optimised for clicks without editorial override, has not made a technical decision. It has made an editorial one.

   Optimisation alignmentEditorial overrideValues consistency

The media variant of the Nexus Ai™ scoring system weights Provenance more heavily than the general framework — because in journalism, source integrity is not just an ethics concern, it is the core professional standard the industry exists to uphold. Representation carries standard weighting but takes on additional specificity: it must account for geographic and geopolitical diversity, not just demographic representation within a single market.

Measurable indicators

Media Ai risk is rarely visible until it has already caused harm — a published hallucination, a copyright challenge, a documented bias audit, an audience trust survey. These ten domains map where that risk accumulates, each tagged to its core principle and monitored across the full lifecycle of editorial Ai deployment.

Editorial Ai ethics isn't a procurement checklist. It is a continuous practice embedded across the full lifecycle — from the moment training data is selected through to how audiences experience content every day. The editorial standards that apply to human journalism apply equally to journalism produced, shaped, or distributed with Ai involvement.

Ethics by design is not just a filter for removing what is wrong. It is a lens for amplifying what is right. The smaller, ethically validated dataset produced more objective outputs, surfaced more representative climate narratives, and made the tool more useful to the journalists it was built for. One tenth of the data. Significantly better results. A product that can be defended — because the ethics work happened by design.

Most media ethics thinking about Ai is happening in the opinion pages, not the editorial governance meetings. We work directly with news organisations, publishers, and media groups to implement the Nexus Ai™ Media Ethics Framework in practice — not as policy documentation, but as working editorial governance embedded in real Ai deployments.