Key Takeaways

  • The $50 million initiative will fund civic AI projects serving up to 100 million residents globally
  • Grants of $500,000 to $2 million will focus on health, mobility, resilience, and civic engagement
  • Applicants need city-government partnerships and credible plans for responsible AI governance

The Rockefeller Foundation and OpenAI Foundation have launched a $50 million program designed to move artificial intelligence from municipal pilot projects into operational public services.

Called “Cities Deliver: AI Solutions for Public Services,” the initiative will support nonprofits and civil-society organizations working with city governments. According to the Rockefeller Foundation, individual grants will range from $500,000 to $2 million, with the overall portfolio intended to reach as many as 100 million residents worldwide.

Applications opened October 14, 2026, and close November 29, 2026. Eligible projects need to involve a city-government partner and address key operational areas such as public health, economic mobility, environmental resilience, or civic engagement.

That structure matters. Municipalities frequently have extensive data and well-defined service problems but limited access to AI engineers, procurement expertise, and governance resources. Nonprofits can provide domain knowledge and implementation capacity, while city agencies contribute operational context and access to public-service workflows.

The program grew partly from a 2023 convening of more than 40 civic leaders from six continents. Participants identified demand not just for AI technology, but also for talent, practical tools, and responsible implementation support. Buying access to a model is the easy part. Integrating it into a benefits office, public-health department, or emergency-response system presents a fundamentally different challenge.

Potential applications are broad. A city could use AI to help residents navigate eligibility requirements for social programs, analyze public-health patterns, translate service information, prioritize infrastructure inspections, or identify neighborhoods facing elevated climate risks. Generative AI could also help public employees search policy documents or draft routine resident communications.

Public-sector deployments carry consequences that ordinary workplace copilots may not. An incorrect summary inside a private document is inconvenient, but an incorrect recommendation affecting housing assistance, health services, or disaster support can materially harm a resident.

That puts governance near the center of the initiative. NIST’s Generative AI Profile identifies 12 risks that are specific to, or amplified by, generative AI. Its approach organizes controls around governance, risk mapping, measurement, and management. City partners could use that structure to evaluate model accuracy, privacy, bias, security, human oversight, and transparency before systems reach residents.

The U.S. Government Accountability Office also offers an AI accountability framework centered on governance, data, performance, and monitoring. For grant recipients, such frameworks can help convert broad responsible-AI promises into concrete practices: documented data sources, testing across demographic groups, appeal channels, audit logs, and clear responsibility when a system produces a harmful result.

Technology choices will vary by project. Organizations may evaluate OpenAI, Microsoft Azure AI, Google Cloud Vertex AI, or combinations of commercial and open technologies. The decision is unlikely to rest on model quality alone. Data residency, integration options, accessibility, language coverage, cost controls, procurement rules, and the ability to change providers could all influence deployment.

Existing public-sector AI work in Cali, Colombia; Cape Town, South Africa; and Rio de Janeiro, Brazil provides useful context for the program. Those examples point toward a global portfolio rather than a grant effort concentrated only in wealthy North American or European municipalities. Local conditions will still shape what works. A resident-service system designed for one city may perform poorly somewhere with different languages, records, connectivity, or administrative processes.

Can $50 million transform urban service delivery on its own? Probably not. It can, however, finance enough substantial implementations to reveal which operating models are repeatable and which remain expensive demonstrations.

The Rockefeller Foundation’s connection to AI stretches back to its investment in the 1956 Dartmouth Conference, where the term artificial intelligence was coined. Seventy years later, the emphasis is less theoretical: measurable service improvements, responsible deployment, and systems that city employees can sustain after grant funding ends. That last part may determine whether Cities Deliver produces durable civic infrastructure or another collection of promising pilots.