The Newsroom That Named Its Terms

A newsroom that puts its AI policy in writing tells its journalists something a newsroom that has not yet done so cannot. It says that when the assistant on the screen misreads a source, misdates a quote, or drops a fabricated paragraph into a running story, the paper will treat the mistake as a decision the paper made. Ownership is what a policy transfers. It is a heavier thing to sign than the vendor contract underneath it. Most of Africa's newsrooms have not signed yet.

One of them has. On 6 March 2026, Nation Media Group in Nairobi launched an in-house Artificial Intelligence framework for its journalists, becoming the first news enterprise in Kenya and the wider East African region to formalise how AI enters the copy desk. The Media Council of Kenya called the move a landmark. NMG's own explainer described the ten principles the framework sets out, from accountability and fairness through human-in-the-loop and explainability, as a working document rather than a wall poster.

What NMG committed to print

The framework's scope is deliberately narrow. It permits AI-assisted translation, transcription, curation, audience analysis, task automation, targeted advertising, and, with restrictions, limited content generation. It requires human review at every editorial hand-off, and it names training in verification and critical thinking as an obligation the group owes its staff. Every clause reads like it was drafted by someone who has watched a chatbot invent a false attribution and had to defend the correction to a reader on the phone. That register, practitioner rather than consultant, is the part of the document worth reading first.

Nation Media Group is the largest newsroom on the continent's east coast. It publishes the Daily Nation in Kenya, runs NTV, and operates bureaus across the region. The March framework places it in the small group of publishers globally, alongside CNN, the New York Times, the BBC, the Guardian, the Financial Times, and Sky News, that have committed a code of AI conduct to print. Most of Africa is not yet on that list, and the question of what its absence costs is the interesting one this month.

Elsewhere on the continent

Read TechCabal's May 2026 write-up of the Carpe Diem Solutions industry report. Nigerian journalists rate AI's impact on their daily work between seven and eight on a ten-point scale. AI-assisted transcription, editing, and drafting have moved from experimental to routine across the seventeen news organisations sampled, from national newspapers through independent digital outlets. The report's second finding, less flattering, is that many of those newsrooms have no formal editorial framework governing that adoption. The tools work. The rules underneath them have not been agreed.

The Reuters Institute Digital News Report 2026 puts Kenya and Nigeria joint first on overall trust in news, at 68 per cent across 48 markets surveyed. The Nigerian country page records the same pattern, alongside concern about deepfakes and AI-generated misinformation running above 75 per cent among readers. This is the paradox a newsroom manager in Lagos or Nairobi lives inside every morning. Readers still trust the paper, and readers also expect the paper to be impersonated by a synthetic version of one of its own bylines inside eighteen months. Every headline is read through that second filter.

The industry association Broadcast Media Africa, in a plenary readout published 29 July 2026, called on African broadcasters to stand up governance frameworks at the level of the operation rather than the whitepaper. Nairobi is on that shorter runway earlier than most.

Cape Town's different bet

Head south, and the register changes. Naspers' Media24 division, which publishes News24 and Netwerk24 in South Africa, is not writing a journalist-facing AI code first. It is instrumenting the sell-side. Media24 launched Match24, a contextual ad-targeting product driven by an AI classifier that scans articles for topic and sentiment, then places ads against emotionally aligned content without cookies or personal data. By 2026 the tool sits alongside audiobook production, translation, transcription, copy-editing tools, and a story-sentiment tracker for brand safety. This is the American ad-tech pattern rebuilt over Afrikaans and English content, and it assumes the productivity lives on the sell-side pipeline rather than in the newsroom.

The Media24 annual results published this year tell you what the transition costs. Revenue fell about 28 per cent year on year, from roughly $141 million to $102 million, as the business moved from print-first to digital-first. Naspers' own explanation is that the drop is a transition line: digital revenue climbing, print falling faster, and the AI investment sitting at the top of the technology bill. That is honest accounting. It is also the paragraph a Kenyan or Nigerian rival will read carefully before committing to the same trajectory, because Media24's balance sheet cushions the fall in a way most of the continent's independents cannot borrow. Nairobi is betting on the newsroom; Cape Town is betting on the sell-side. Both readings are defensible. Neither is obviously right.

The chapter the framework leaves unwritten

Now the disconfirming source, and the one that keeps me up. The Media Diversity Institute's essay on AI in African newsrooms puts the register on the labour that trained the models in the first place. The same tools making a Lagos reporter faster on transcription and a Nairobi sub-editor faster on translation were partly built on Kenyan and Nigerian annotation labour subcontracted through firms like Sama, working on rates the market says nothing about in a Meta, OpenAI, or Anthropic quarterly release. The productivity is not free. Its cost sits on another continent's P&L, or off any P&L at all, in a labour category that shows up in nobody's newsroom KPI dashboard.

Answering the essay rather than strawmanning it: the point is not that Kenyan newsrooms should refuse the tool. It is that any newsroom AI policy written in 2026 that does not name the annotation supply chain, and does not commit to reviewing where its vendors got their training data, has left a chapter unwritten. Nation Media Group's framework, as published, is silent on that chapter. So is every peer document I have read. The gap is the honest reading of where the industry sits. It is not a knock on Nairobi's document specifically. It is a knock on the whole cohort.

A pragmatist's paper, and where training lands

The strongest academic reading of where African journalism actually sits with AI is the April 2025 Journalism Practice paper by Umejei, Ayisi, Phiri, and Tallam, which interviewed reporters and editors across Nigeria, Ghana, Kenya, and South Africa. The finding is that African newsroom workers cluster into three postures — optimists, pessimists, pragmatists — with pragmatists dominant in daily practice. Pragmatism, the authors argue, means using AI where it plainly helps, distrusting it where it plainly does not, and taking the trade-off on the copy in front of you rather than on a whitepaper. That is the register the NMG framework tries to codify in an institutional voice. It is a more useful starting point than the optimist-pessimist axis a lot of Western reporting still leans on.

I ran a small workshop in Kampala in 2019 for editors and data reporters from three East African newsrooms on data-driven investigation. Half the room used pivot tables like a scalpel. The other half had never opened one. The people who produced the best work were the ones who understood the tool's limits before its features. Nothing about the arrival of a generative model changes that ranking. The NMG framework, at its plainest, is an attempt to describe the tool's limits before its features. That is why it will do more editorial work in the first year than compliance work.

One line of the framework matters more than the other nine. NMG commits to staff training in AI and data literacy, critical thinking, fact-checking, verification, and ethical considerations. That clause is where any productivity uplift actually lives. Without training, the tool operates on the journalist rather than the other way round. Every JournalismAI cohort readout I have seen out of Sub-Saharan Africa reports the same shape: trust in the tool runs ahead of training on the tool. Training is the underinvestment that decides whether AI in the copy desk is uplift or subsidence. The practical question for every publisher watching Kimathi Street is not whether to write a framework this year. It is whether the training line inside the framework is funded to the level of the vendor licence. Where those two lines match, the paper has staffed the shift honestly. Where the licence is paid and the training is not, the paper has bought a tool and outsourced its judgment.

Kimathi Street, Wednesday evening

There is a corner desk on the third floor of Nation Centre, the building on Kimathi Street where the Daily Nation is put to bed each night. In the second week of March, a sub-editor sat there and read her paper's AI framework in printed form for the first time, before it went live on the intranet. She was old enough to remember the newsroom before Twitter. On her screen, a transcription tool was churning through the tape of a courtroom hearing she needed a summary of by ten. On paper, in her hand, was the document that would decide what she was permitted to publish from that summary tomorrow. Neither the tool nor the framework was going to write the story for her. Both were going to be part of what the paper filed on Kimathi Street the next morning. She read the framework once, folded the pages in half, and got back to the transcript.

Tarry Singh is the founder and CEO of Real AI, an enterprise AI advisory and deployment firm working with global enterprises on production agent systems, model risk, and AI sovereignty strategy. He also leads Earthscan for Energy AI startup, and is a founding contributor to the EU-funded HCAIM and PANORAIMA programmes for responsible AI education across European universities. He writes at tarrysingh.com.