Credits are not infrastructure

The release is one page. The sentence that matters comes a few paragraphs down: every participating newsroom "will also receive credits to OpenAI's API to facilitate the development of bespoke newsroom solutions." There is no number in that sentence. Everything I want to say about this programme starts from the missing number.

To the release's credit, it is better than most. It does not promise a transformation and it stays away from the vocabulary these announcements usually arrive in. It describes six webinars, twelve weeks of coaching, ten newsrooms and some credits, and stops. Modest, specific, honestly worded. And still, read as an engineer, it hands a metered dependency to a sector whose own research says its problem is money and people, and calls that support.

What is actually on offer

The programme has two phases. The first was a six-part webinar series, 5 to 28 August, open to 40 local and regional newsrooms across Ukraine. The second starts on 17 September: twelve weeks of coaching for ten selected newsrooms, with an in-person kick-off in Paris after the Paris AI Forum and a virtual showcase on 15 December. Each newsroom is meant to leave with "an AI workflow or prototype, alongside a clear strategic roadmap". Plus the credits. The amount is not stated for Ukraine. For scale, the global fund WAN-IFRA and OpenAI set up in April 2025 was $1.5 million in API credits, spread across 128 newsrooms in four regions.

Oksana Brovko, who runs AIRPPU, says the programme "should give journalists back time to do journalism." I take that at face value. It is a reasonable goal and a modest one. The question is whether the programme matches the problem. The sector's own diagnosis

In July, the Media Development Foundation and Internews International presented their ninth annual study of Ukraine's regional media, funded by the EU, based on 61 newsroom questionnaires and ten interviews with managers. It is the closest thing to an independent baseline for the newsrooms this programme is aimed at, and nobody in it has a stake in OpenAI.

The headline finding is that in 2025 "lack of funding became the sector's top challenge, overtaking staff shortages": 72 percent of editors named it among their three most urgent problems, up from 58 percent a year earlier. The suspension of USAID programmes at the start of 2025 "affected the vast majority of surveyed outlets". Staff is the second problem, and for the first time in the study's history, mobilisation is the leading reason people leave a newsroom, reported by 15 of the 61 outlets. One editor in central Ukraine said the risk of mobilising male team members "casts a huge shadow over the future of media" and is "an existential threat to the sector."

The study's recommendation to donors is to "shift from short-term crisis funding to more sustainable, flexible support models." A twelve-week accelerator with an unspecified voucher is, in the study's own vocabulary, short-term support. "Give journalists back time" is a productivity claim, and productivity claims help organisations that have people. A transcription workflow does not replace the reporter who was mobilised, and it does not pay the one who stayed.

Credits are a countdown

Here is what "bespoke newsroom solutions" built on API credits looks like in practice. A thin application sits in front of a hosted American model that charges per token. For the length of the programme the tokens are free. On the day the credits run out, the newsroom owns a running cost, in dollars, in a market where 67 percent of outlets describe regional advertising as "very weak" or "virtually absent." The fashionable reading is that building your own tools is independence. McKinsey's 2026 survey, published in August, is already being quoted for it: 32 percent of respondents report their organisation "decided against buying one or more software products or features because they could be built internally with agentic coding tools." Self-reported, one product or more, concentrated in technology and healthcare, and beside the point here. Building your own front end on somebody else's metered API is not independence. You now own the code and rent the brain.

I know how this feels from the inside, because I am two months into the same arrangement. I maintain an open-source library, four years old, around 1,300 commits. In July I applied for Anthropic's programme for open-source contributors, which gives maintainers its top Claude subscription free for six months, and I was accepted. Since August the model has been a main contributor to the project. In February the free period ends and the subscription becomes my cost, at whatever the price is by then. That is exactly what the credits are for, and it is exactly the problem.

This is not the first time platforms have paid newsrooms to adopt their technology. The Tow Center at Columbia published a report in May 2025 on a decade of Google and Meta money in journalism. Its summary of tech companies' efforts to engage with news over that period: "inadequate financial support that arrived sporadically and was tied to conditions," after which the two companies went "to reducing their focus on journalism (Google) or withdrawing entirely (Facebook) in the space of a couple of years." One news executive told the researchers that "news, as an industry, tends not to learn terribly well from our past experiences and mistakes." Another said that on partnering with platforms, "I would have more follow-up questions than a younger me did." That report was about search and social. The credits model has the same shape with a meter attached.

Who pays for the diagnosis

One European note, because it is sitting in plain sight. The study that documents what Ukrainian regional newsrooms actually need was paid for by the EU. The tool on offer is a voucher for a US API. The same week this programme was announced, South Korea picked consortia led by SK Telecom, KT and Kakao to build a free national AI service with domestic models required for at least half of it, explicitly to reduce dependence on foreign platforms; it is in beta now, with a full launch expected by the end of 2026 and state subsidy from 2027. I am not arguing that Ukrainian newsrooms should wait for a European model. But Europe funded the research and left the tooling to OpenAI.

The obvious objection

The strongest answer to all of this is that the programme never claimed to be infrastructure. WAN-IFRA has run the Catalyst since 2024, across 128 newsrooms, and its CEO Stig Ørskov says the OpenAI collaboration "has enabled publishers worldwide to test, question and shape how AI is used in news." Testing costs money, credits make testing free, and a newsroom that learns in twelve weeks that a transcription workflow is not worth its token bill has learned something useful at no cost. On that reading I am criticising a pilot for not being a budget.

I accept that for the webinars. I do not accept it for the deliverable. The programme's own page says each newsroom leaves with "an AI workflow or prototype" and a roadmap for "the next steps of their AI journey." A roadmap is a commitment to keep going. If it does not carry a cost line for the day the credits stop, it is a roadmap into a bill, and nobody at the December showcase will be the one paying it.

What I want to see in December

Not ten prototypes. Ten monthly token bills, each with a line for who pays it in January. If those numbers are small, the programme worked and the critics, me included, should say so in public. If they are not, WAN-IFRA and OpenAI will have taught ten newsrooms to build something they cannot afford to run. The sector has had that lesson before.