I want to start with a confession: I did not go to Bellagio to think about AI.

I went because the Aspen Institute had convened something genuinely rare: 16 bipartisan, bicameral Members of Congress, national security scholars, AI researchers, and foundation colleagues at the Rockefeller Foundation's Bellagio Center in Italy, spending five full days together entirely outside of Washington.

As someone who works on congressional engagement for global health and nutrition policy, I saw that room as the point. The relationships, the candor, the conversations that don't happen in a 30-minute Hill meeting.

But when I came home, all I could think about was AI.

Not because the conversations were alarming, though some of what I heard was sobering. But because, somewhere in those five days, I realized that the foreign assistance community is navigating an enormous policy inflection point at exactly the moment when AI is changing what effective programming and advocacy look like. And we are not fully prepared for either.

This piece is about what I learned, what I'm actually doing about it, and most importantly, what I think government decision-makers and those who engage with them need to consider right now.

The Missing Perspective

At Bellagio, I was something of an odd fit — representing the Eleanor Crook Foundation (ECF), a philanthropy focused on ending childhood malnutrition through funding, programming, and advocacy. 

Malnutrition and foreign assistance more broadly may seem far removed from AI governance debates. But in practice, it means working on the systems that determine whether people can access basic institutional support amid strain.

Malnutrition and foreign assistance more broadly may seem far removed from AI governance debates. But in practice, it means working on the systems that determine whether people can access food, healthcare, public services, and basic institutional support amid political and economic strain.

Many global health organizations operate in places where state capacity is weakest, infrastructure is fragmented, and institutional failures carry immediate human consequences. Those are exactly the environments where the risks and governance gaps around AI are likely to surface first.

Most AI governance conversations still focus on the actors building the systems and the states competing to control them. Those issues matter. But they are not the only systems AI will reshape.

The same communities most likely to be excluded from AI’s benefits are often the same communities global health organizations have spent decades trying to reach

The same communities most likely to be excluded from AI’s benefits are often the same communities global health organizations have spent decades trying to reach: populations facing weak infrastructure, language barriers, workforce shortages, and limited public-sector capacity.

AI Is Reshaping Public Health Systems

Many organizations are already deploying AI systems across diagnostics, screening, evidence synthesis, and public health operations. We can learn from them. 

In diagnostics

D-Tree International has spent a decade building digital decision-support tools for community health workers. The next generation integrates AI to improve accuracy and adapt to local conditions. For severe acute malnutrition (SAM) — which can kill a child within weeks and is massively underdiagnosed — AI-assisted screening is being piloted to help community health workers identify cases earlier without specialist backup. 

Qure ai is deploying AI-powered chest X-ray analysis for tuberculosis screening across India, Nigeria, and other high-burden settings, achieving diagnostic accuracy that matches specialist radiologists at a fraction of the cost and at a scale the health workforce alone cannot achieve.

In evidence synthesis

Institutions like the International Food Policy Research Institute are using AI to accelerate systematic reviews — the evidence documents that underpin clinical guidelines and policy recommendations. For nutrition, where the evidence base is deep but scattered across decades of research in multiple languages, this matters: a systematic review that once took 18 months can be meaningfully accelerated, which matters when policy windows open and close faster than traditional research timelines allow.

The labor and health connection

The International Labour Organization's 2023 analysis of generative AI and jobs found that low-income countries are likely to benefit least from AI's productivity gains, because they lack the complementary infrastructure — connectivity, trained workforces, regulatory capacity — that allows AI tools to be deployed effectively. 

Global health advocates need to be making that connection for policymakers who currently see AI policy and health policy as entirely separate domains.

The downstream health consequences of AI-driven economic disruption in fragile contexts — through income loss, food insecurity, reduced access to care — are not yet well modeled. Global health advocates need to be making that connection for policymakers who currently see AI policy and health policy as entirely separate domains.

These systems may improve reach and efficiency. But they also introduce new governance questions.

• Who owns the underlying data?
• Who audits model performance across different populations and languages?
• What happens when donor-funded pilots end?
• Which governments have the technical capacity to evaluate vendor claims before deployment?
• Who becomes dependent on proprietary infrastructure that they cannot meaningfully govern?

AI Is Changing Who Can Participate Effectively in Policymaking

AI is also changing advocacy itself.

Organizations with access to advanced AI tools can now generate tailored legislative analysis, synthesize evidence, model counterarguments, and adapt communications at speeds that smaller civil society organizations often cannot match.

That creates a new form of institutional asymmetry.

In my own work, I increasingly use AI to prepare highly tailored congressional briefing materials based on committee assignments, district priorities, prior voting records, and public statements. 

In my own work, I increasingly use AI to prepare highly tailored congressional briefing materials based on committee assignments, district priorities, prior voting records, and public statements. I use it to stress-test arguments from multiple ideological perspectives before difficult meetings. I use it to translate dense nutrition and development evidence into language policymakers can absorb quickly.

Building tailored briefs for congressional meetings

A generic one-pager no longer cuts through in most congressional offices. AI helps me build meeting-specific briefs calibrated to the individual member: their committee assignment, prior votes on foreign assistance, their district's economic ties to global health programs, and their stated priorities. AI handles the research and first draft; I focus on accuracy and argument. The result is a document that signals genuine knowledge of the member's context, which changes the tone of the conversation.

Stress-testing messages before they go into the room

I use AI as an adversarial thinking partner: I'll draft a framing for a difficult argument — say, why continued investment in the World Food Programme's nutrition programs makes sense in a fiscal austerity environment — and ask AI to push back from multiple ideological positions. The objections it surfaces are not always ones I'd identify on my own. Working through them makes the real conversation sharper.

Translating evidence for non-specialist audiences

The return on investment for treating SAM with ready-to-use therapeutic food (RUTF) is one of the most compelling numbers in global health — the Copenhagen Consensus has consistently ranked nutrition interventions among the highest-return development investments. But that evidence resides in technical literature inaccessible to most policymakers without translation. 

Along with recent polls conducted by organizations like the Kaiser Family Foundation and even McLaughlin & Associates, I use AI to extract the two sentences a member of Congress actually needs to hear, in language that connects with their priorities. That is the core work of advocacy, and AI makes it faster.

Vocal about the Access Gaps 

Technology helps. AI may lower some barriers to expertise while simultaneously widening capacity gaps between well-resourced institutions and smaller organizations working most closely with affected communities.

The efficiency gains I’m describing accrue primarily to advocates who operate in English, have stable internet access, software budgets, and the institutional capacity to integrate these tools into daily workflows.

Most AI debates still assume digital literacy, reliable connectivity, strong regulatory capacity, and governments capable of implementing sophisticated oversight systems. Yet much of the world does not operate under those conditions.

Most AI debates still assume digital literacy, reliable connectivity, strong regulatory capacity, and governments capable of implementing sophisticated oversight systems. Yet much of the world does not operate under those conditions.

That creates a serious imbalance. The organizations most embedded in frontline realities are often the least equipped to compete in this new informational environment. That is not a reason to stop using AI. It is a reason to be vocal about the access gap and to advocate for the resources that would allow local organizations to benefit as well

The Train Is Moving. Help Steer It.

I came back from Bellagio with more conviction than when I left, not only about the future of AI, but also about the importance of foreign-assistance advocates being present in the spaces where AI and power are being shaped.

The frameworks being designed now by governments, philanthropies, multilateral institutions, and private companies will dictate how health systems allocate resources, interpret evidence, deliver services, and interact with citizens for years to come.

The communities whose lives depend on the decisions being made now—about what global health investment looks like, about how AI-enabled health tools are deployed and to whom—cannot afford for us to be absent. 

The good news is that the members of Congress I spent time with are genuinely open to being persuaded by evidence and expertise. The philanthropic community I help steer is actively looking for where to put resources. 

The good news is that the members of Congress I spent time with are genuinely open to being persuaded by evidence and expertise. The philanthropic community I help steer is actively looking for where to put resources. The global multilateral institutions are looking for partners.

The work is showing up, being specific, and being useful. We have better tools to do that than we've ever had before.