A finance ministry official can now draft in an afternoon what used to take a WHO mission, a consultant and a wait of weeks. That single fact, multiplied across almost everything WHO produces, is the real story behind this year’s budget cuts — and almost nobody in Geneva is telling it yet.

Picture a health ministry official in a mid-income country. Two years ago, adapting a WHO clinical guideline into a national protocol meant WHO staff time, a consultant, or waiting for the next country mission.

Today she opens a general-purpose AI assistant, points it at WHO’s own – largely open-access – guidance library, and has a serviceable first draft before lunch. Nobody announced this. No governing body voted on it. It simply became true, quietly, over the past two years, and it is already reshaping what happens when member states next debate how much to pay Geneva.

That is the story beneath WHO’s budget cuts, and WHO has not yet told it in public.

New rulebook – and the harder question underneath it

Last month, at the AI for Good Global Summit in Geneva, WHO joined the International Telecommunication Union and the World Intellectual Property Organization to launch a joint framework on AI in health innovation timed almost exactly to a moment when generative-AI patents published over the previous two years overtook the total from the entire preceding decade. 

It extends a six-year pattern of WHO writing, with real skill, the rules by which the world should govern AI. What none of that writing addresses is what AI does to the value of what WHO itself is paid to produce.

Bolting an AI layer onto business as usual – a copilot here, a chatbot there – will not answer that. What is arriving is a change in what WHO’s outputs are worth, who can produce them, and what member states believe they are buying when they pay their dues. The window for choosing WHO’s place in that shift is the term of the next Director-General, not some comfortable decade beyond.

Two shocks, one organisation

Two shocks are landing on WHO at once, and so far only one has been reckoned with in public.

The first is financial. In February 2025, the Executive Board cut the proposed base budget for 2026–27 from $ 5.3 billion to $ 4.9 billion. By May, the Assembly had approved US$ 4.267 billion – a 9% cut on 2024–25, 22% below the original ambition. 

In the same vote, member states approved a second consecutive 20% rise in assessed contributions – the fixed dues every government owes – continuing a path toward those dues covering half of WHO’s base budget by 2030–31. 

Why governments would vote to send Geneva more fixed money in the very years an AI assistant is learning to produce what those dues used to buy is a question this piece returns to.

The response has been a formal prioritization and realignment exercise, still running. WHO’s global workforce stood at 9,473 in July 2024 and 9,457 in December 2024, a 15-year peak by WHO’s own account. A WHO report obtained by Health Policy Watch projected roughly 2,371 separations against that base, implying a mid-2026 total near 7,086, this author’s arithmetic, not a published figure. The latest actual snapshot (31 December 2025) put headcount at 8,569 – already 888 below December 2024.

Geneva headquarters, WHO’s largest office, is on track to shrink 28% by mid-2026, Africa and Europe close behind at 25% and 24%. Even after those cuts, WHO faces a $141 million gap in 2025 salary costs, and a projected $1.05 billion funding gap for 2026–27, down from $ 1.7 billion estimated in May 2025.

This is an unusually well-documented contraction, which is exactly why it is useful: it gives a checkable baseline against which the second, technological shock can be measured rather than guessed at.

Two workforces, one falling cost base

WHO’s people split into two legal populations. Established staff – 8,569 as of 31 December 2025 – are one. Affiliates – consultants, Special Service Agreement (SSA) holders, and Agreements-for-Performance-of-Work (APW) holders – are the second, and this group is contracting fastest: 9,937 cumulative engagements in 2025, down from 12,965 the year before. Comparing matching windows, SSA holders fell 21.4% year-on-year, consultants 22% in headcount and 25.5% in full-time-equivalent (FTE) terms, APW holders 31.8% in headcount and 34.2% in FTE terms.

This cannot be pinned on AI alone. Affiliate contracts are the fastest lever any organisation has for cutting cost. What can fairly be said is the contraction is real, twice-measured, and concentrated in exactly the deliverable-based, language- and data-centric work this piece flags as most exposed to automation.

Senior management has been reshaped rather than thinned. Between 2017 and 2025, P6 posts fell 42% while D1 and D2 posts rose 29% and 31%; net across senior grades, a 9% reduction – even as entry-to-mid P1–P3 posts faced a projected 30% cut. Whether AI accelerates or repeats that pattern through 2030 is genuinely open.

The cost base is precisely known. As of the January–July 2024 half-year, staff costs were $814 million – 47% of total expenditure, up from 36% a year earlier. For 2026–27, a staffing paper reviewed by Health Policy Watch put projected total staff-related costs at $2.26 billion, of which $1.19 billion (52%) is contractual services or consultants.

Not all WHO work is the same

Here the story turns from documented fact to informed argument. WHO does at least five distinguishable kinds of work, unevenly exposed to automation. 

The ranking below is my own qualitative framework – low, medium or high, no percentage attached, because none is measured – anchored in two much-cited studies finding writers, translators, analysts and clerical occupations among the most exposed to large language models. Document production, translation, data processing and analysis are precisely what a large share of WHO’s staff and consultant time buys.

A claim making the rounds in Geneva – that AI will “replace 80–90% of WHO jobs” – is both true and false, and the gap is the point. That figure is this author’s own working assumption, set above the published research’s central estimates. 

AI is plausibly on track to automate most tasks filling professional staff and consultant time today. It is nowhere near replacing the functions that justify WHO’s existence. The danger is that member states, watching the first happen in plain sight, quietly stop paying for the second.

Call the mechanism task hollowing: each role loses most of its routine content, headcount needed per output falls, and the humans who remain concentrate in judgement and accountability — consistent with the ILO’s own conclusion that transformation, not disappearance, is generative AI’s most likely impact. 

The affiliate contraction and the P6-to-D1/D2 reshaping above are both consistent with this happening inside WHO now, though neither can be pinned on AI specifically rather than budget pressure alone.

The shift is audible in donor language. At last month’s summit, the Global Fund’s John Fairhurst told a panel that countries want efficiency, more impact per dollar, and AI is the pathway they are reaching for. That is a financier approvingly describing exactly the substitution mechanism this piece warns about.

Three phases, and the trap inside the savings

Three phases seem likely to 2030 – my scenario, not a WHO projection: an assistive phase through 2026–27, where staff use AI individually with little structural headcount change; an agentic phase, roughly 2027–29, where agents own whole workflows and affiliate and admin posts contract further; and a substitution phase from around 2029, where member states run their own AI health-intelligence capacity and WHO’s value as output producer approaches zero.

The variable that actually decides WHO’s financing is not how busy its staff are, but how much member states value what only WHO can provide. Countries will not stop funding WHO because its staff stop working. They will stop to the extent an AI assistant hands them, for a fraction of the cost, the report they used to rely on WHO to produce. The question shifts from “does WHO work hard?” to “what can only WHO do?”

But there is a trap inside the savings this implies. Applying illustrative cut rates to the two verified cost figures above – from a cautious 25%/10% to an aggressive 55%/30% – yields plausible annual savings by 2030 of roughly $202 million to $488 million: bookends built on two real numbers, not a forecast. 

But every franc WHO saves by automating production is a franc it has just demonstrated a member state could save at home. Savings are necessary. They are not a strategy.

The member state question

Every financing conversation about WHO has so far assumed the only variable was generosity. AI changes the question itself: it is no longer only about willingness to pay, but whether the thing being paid for still needs to be bought from Geneva at all.

The timing of WHO’s own financing reform makes this uncomfortable. In 2020–21, assessed contributions covered just 16% of WHO’s base budget; in 2022 the Assembly agreed to raise that to 50% by 2030–31, and member states have since approved two consecutive 20% increases. Governments have voted twice to send WHO more fixed dues in exactly the years an AI assistant is becoming capable of producing much of what those dues used to buy.

WHO’s own investment case claims every dollar invested delivers a return of at least $35 – logic that depends on WHO being the necessary producer of the goods being valued. 

To the extent a ministry can generate the report itself, that return has to be recalculated, not because WHO got worse, but because the alternative got cheap. Before the next dues vote, member states will quietly ask: what does WHO supply that we could not now generate ourselves? Where the honest answer is “not much,” that dollar is at risk.

There is a genuine counter-argument, and it is WHO’s strongest card. A world of 194 finance ministries each generating their own AI-assisted guidance, with no shared quality bar, is a world of fragmented, occasionally wrong health advice – exactly what a global normative body exists to prevent. 

The risk is not speculative. Alain Labrique, WHO’s director of data, digital health, analytics and AI, warned at the summit that imported models are typically trained on data unrepresentative of the people they are meant to serve. This is echoed by Harvard researchers who note roughly 90% of global genomic data belongs to people of European descent.

The honest answer is not that WHO should out-produce the AI – it will lose that race on cost – but that WHO should become the body that certifies whether the AI got it right, anchored in convening authority no single ministry can replicate. 

As HealthAI’s Ricardo Baptista Leite put it in Geneva: “Innovation moves at the speed of trust.” Trust is the one input WHO can still supply more cheaply than anyone, which makes it strange how slowly Geneva has moved to industrialise exactly that.

Writing the rules for the world

WHO has, to its credit, been an early and prolific author of AI governance for the world. Set that against what the same institution was doing with AI inside its own walls, and the contrast is hard to miss. The HR process document governing the current restructuring describes a wholly manual sequence of spreadsheets and hand-built organigrams, with no visible role for the AI tools whose ethics WHO was simultaneously instructing the world how to govern.

The UN system tells a version of the same story. Secretary-General Antonio Guterres launched the UN80 reform initiative in March 2025 but independent analysis in December 2025 found no formal mechanism yet existed to advance AI proposals system-wide. It took until January 2026 for the UN to announce its first system-wide staff AI-literacy partnership – roughly three years after ChatGPT’s public release.

This is not an outside critic’s complaint. Anders Nordström, WHO’s former acting Director-General, made close to the same argument in Think Global Health last month, reclaiming trust requires WHO to become excellent at what it alone can do. “Modernization must begin internally,” he concluded.

None of this is entirely unreasonable. Rules built for a different era are not obviously wrong to apply cautiously to a technology prone to fabricating plausible text. But caution and speed are not opposite ends of one axis: banking supervisors and hospital systems deployed internal AI copilots under equally strict rules well inside the three-year window it took the UN system to reach the starting line.

Four futures

Two independent choices – how fast AI is adopted, and whether WHO repositions from producer to steward – generate four outcomes: a discussion framework, not a model with predictive weights.

Only one quadrant is durable: fast adoption paired with a genuine shift to steward. Fast adoption without repositioning gives credibility without capability. Slow adoption while staying a producer is managed decline. 

Slow adoption with continued cost-cutting is the sharpest trap – implosion by efficiency, where every saving proves to member states they could have made it themselves, funding falls further, capable staff leave, and WHO shrinks into a smaller producer of outputs that matter less each year.

The inheritance, and what has to happen now

A new Director-General takes office in 2027, inheriting a financing contraction already locked in at $4.267 billion, and a technological displacement that is only just beginning. 

Stabilisation is the wrong objective: rebuilding the old equilibrium means rebuilding an institution optimised for a world in which technical outputs were expensive to produce – a world that is ending. The next Director-General will be remembered either as the leader who managed WHO’s decline with dignity, or who repositioned it for the AI era. There is no third option.

For incoming leadership: name the producer-to-steward shift on day one; write the rules for AI in health before regulators and private platforms fill the vacuum; decentralise deliberately rather than by budget accident; extend the current restructuring machinery to affiliates, currently outside its protections; and re-contract with member states around value, not volume. 

Nordström’s own reform agenda reaches a similar place from a different direction – sharper mandate, single-term leadership, open recruitment of regional directors, a functioning board of trustees, financing primarily through assessed contributions. Independent reform voices and this piece’s automation argument converge from separate directions on the same conclusion: the institution that survives is smaller, more disciplined, and clearer about what only it can do.

For WHO management: reposition the value proposition before the savings, not after; treat AI adoption as core infrastructure, not a side-project; manage the transition as workforce transformation, not attrition; govern the affiliate workforce deliberately, since it has no continuing contract and no comparable safety net; and publish WHO’s own analysis of AI’s workforce impact before an outside body, or this piece, becomes the only source anyone can cite.

For member states: fund the transformation, not just the contraction; be explicit about what you are buying – norms, trust and equity, on their own terms; protect the global-public-good core; and use the 2027 transition to mandate the producer-to-steward shift, then measure the next Director-General against it.

For staff: the next shock is structural, not cyclical – plan a career on the assumption that routine task-content does not come back, and the transformed job rewards judgement over throughput; demand transformation governance, not just consultation on cuts; and insist savings are reinvested, not banked.

Warning that does not stop at WHO’s door

The pattern is not WHO-specific. On 1 December 2025, Guterres presented the UN Secretariat’s revised 2026 budget: a $577 million (15.1%) cut, and a reduction of 2,681 posts – 18.8% of the Secretariat’s regular-budget staffing table.

Every knowledge-intensive public institution that defines its value by the outputs it produces is exposed as those outputs become nearly free to generate. The institutions that last will be the ones whose value rests on what AI cannot supply: legitimacy, convening power, trust, accountability, and stewardship of public goods no single actor can be trusted to hold alone.

WHO will not be destroyed by artificial intelligence. It can only be destroyed by failing to understand what artificial intelligence makes it for. The task facing this generation of leadership, staff and member states is to make sure that when the cost of producing health knowledge falls to nearly nothing, the world still understands why it needs a World Health Organization — and chooses, deliberately, to keep funding one.

That ministry official, drafting her national protocol in an afternoon, is not the enemy of that outcome. She is the earliest, clearest evidence of the question WHO now has to answer: not whether it works hard enough, but what, in a world where a laptop can draft almost anything, only WHO can still be trusted to do.

A note on the numbers: Verifiable figures above — headcounts, budgets, contribution shares, contract counts — come from WHO’s own governing-body documents, published HR tables, or named reporting. Everything forward-looking — the 2030 scenario lines, the three-phase trajectory, the savings ranges, the four-futures framework, the automation-exposure rankings — is this author’s own analytical synthesis, built on those verified figures but not a WHO forecast. Where automation is described as consistent with the affiliate contraction or senior-grade reshaping, that is interpretation, not proven causation — budget pressure alone could produce the same numbers.

K. Rifat Hossain is Health Policy Watch’s Director of Development. He worked for the WHO for nearly two decades at headquarters in Geneva, the Regional Office for the Western Pacific and the WHO Country Office in Poland, working on health data and intelligence, refugee and migration health, and digital and AI systems for health. He has also worked for the ILO and the UN Economic Commission for Europe.

Image Credits: AI generated by picai.