How Civil Society Groups Are Using Cloudflare to Build AI Tools

For civil society organizations, AI’s practical promise is less about replacing specialist work than clearing away some of its manual steps. Cloudflare says 30 organizations in its first nonprofit startup cohort are using its developer services to build tools for journalism, mental health support, community mapping and human rights research. The company provided more than $7.5 million in service credits across the program.

The projects show how groups are moving from asking technology providers to secure existing services to building software tailored to their own work. They also illustrate the conditions that shape adoption: sensitive information, limited budgets, small technical teams and a need for people to check consequential outputs.

Security and cost shape what teams can build

Cloudflare’s support for civil society predates the startup cohort. Its Project Galileo offers free cybersecurity services and protects more than 3,400 domains across more than 120 countries. The company says organizations that work in the public interest have increasingly asked for help creating applications as well as protecting the ones they already use.

That shift comes as many organizations report financial strain, with reductions in public funding linked to staff cuts and program closures. Some groups are turning to AI to do more with fewer resources, but adoption remains uneven. In a CIVICUS survey, more than half of respondents cited privacy as a barrier to AI use, while 48% said financial constraints limited adoption.

Those concerns are especially pointed for organizations holding information such as activists’ locations or anonymous sources’ identities. Cloudflare says its own data shows that civil society groups face disproportionate cyberattack risk. A tool that saves staff time can still create new exposure if sensitive records are sent to systems without adequate safeguards, or if generated content is accepted without review.

Cloudflare’s developer services are designed to address some of these constraints. Applications running on its network receive protection against distributed denial-of-service attacks and attempts to access internal systems without authorization. Its serverless architecture scales with demand, so teams don’t need to provision infrastructure that may sit idle. Workers AI removes the need to operate dedicated GPU infrastructure, while AI Gateway offers rate limits, usage monitoring and model routing intended to reduce unnecessary calls and manage costs.

These features can lower the operational burden, but they don’t settle questions about whether a model is appropriate for a task, what data it should process or when a person must make the final call. The projects Cloudflare describes offer examples of teams working through those choices in specific settings.

Tools for local reporting, counseling and recreation

LebTown, a nonprofit newsroom covering Lebanon County, Pennsylvania, is replacing a scattered workflow built around two Google Docs, Google Calendar, Discord and Gmail. Its custom editorial system tracks stories from pitch to publication, records audience reach and lets reporters log community impact through the app or Discord. The team has also built software that turns county property transfer records exported as PDFs into structured data, plus a live copyediting agent connected to its WordPress site.

The appeal here is integration: reporting, editing and impact tracking can happen in one workflow, while repetitive document processing becomes easier to manage. The source doesn’t report accuracy results or how much time the newsroom has saved, so the practical value will depend on how well the tools fit reporters’ existing routines.

Kaya Guides uses WhatsApp to connect people experiencing depression in India with trained lay counselors, offering support without the cost or waitlists associated with traditional therapy. As of September 2026, the service had supported 10,439 people, with 2,325 enrolled at that time. It now processes around 500,000 WhatsApp messages each month after moving its care management system to Cloudflare.

The organization has also built an AI system that listens to counseling calls and gives counselors feedback during sessions. Kaya says this has helped it make its program more consistent and self-correcting as it grows, without significant added cost or complexity. The account doesn’t specify how the system measures session quality or how counselors handle feedback when it conflicts with their judgment. Those details matter for a tool operating in a sensitive care setting.

The Snorkelling Society, a UK nonprofit, is developing SnorkelMap, a web and mobile service for finding snorkeling locations and sharing information about them. Volunteers build and maintain the application. Cloudflare provides secure storage for community contributions and uploaded images, alongside protection against cyberattacks. In this case, the main role of the infrastructure is to support a community resource that a volunteer team can maintain.

Automating research while keeping people involved

For larger organizations, Cloudflare describes a recurring challenge: information collected manually from many sources must be brought together, reviewed and structured before researchers can analyze it. That material can include the names of people harmed by human rights abuses. Unauthorized access, leaks or fabricated model output could have serious consequences.

Cloudflare provided developer services and a team of volunteer engineers, including product specialists, designers, builders and security advisers, to help create tools for three organizations. The work is exploring which tasks benefit from automation, which models suit those tasks, what safeguards are needed and where a person needs to stay involved.

Freedom House maintains a database of incidents in which states reach across borders to silence people in diaspora and exile communities. Identifying relevant cases and coding patterns across thousands of potential incidents takes substantial staff time. The team is building an interactive dashboard to collect public reporting, structure it and flag potential cases. It is being fine-tuned to follow Freedom House’s research method, with the organization involved throughout development. The project prioritizes security and data minimization. Its intended role is to help researchers with the initial screening stage, leaving them more time for deeper analysis and reports.

The Global Network Initiative (GNI), a membership organization focused on expression and privacy online, is developing a dashboard to identify developments across dozens of jurisdictions. Its scope includes online transparency requirements in the European Union and data localization mandates in the Asia Pacific region. The system surfaces and categorizes relevant news, requests for comment and legislative activity, then recommends potential advocacy opportunities based on GNI’s mission and previous work. It also tracks internal review and flags approaching deadlines.

Article One advises companies on the human rights effects of their policies, products and operations. Its assessments can require staff to read and summarize material ranging from factory audits to country context reports. The team is working on a tool that structures those documents, identifies significant risks and drafts recommendations for staff to review and refine. Article One retains responsibility for its judgments; AI handles information organization and initial drafting.

Across the three research projects, the intended benefit is a faster path from dispersed material to something a specialist can evaluate. The boundary is equally clear: these systems can help identify, arrange and draft information, while trained staff remain responsible for interpreting evidence and making decisions.

A second cohort is opening

Cloudflare says the first cohort demonstrates how automation can help community service organizations expand their work while keeping infrastructure needs and costs manageable. The company also sees direct collaboration as a way to learn about security and privacy needs in high-risk settings. It has opened applications for a second nonprofit cohort through its startup program; applicants can select the nonprofit option.

The first set of projects suggests that useful AI adoption depends on fitting tools to existing work and defining where human review belongs. The next measure to watch is how these systems perform in routine use: whether they reduce staff workload, maintain accuracy and protect the information their organizations are trusted to handle.