The Foundation Layer

Overview

The Foundation Layer is a philanthropic guide and donor-advised fund created by Tyler M. John, who leads AI work at the Effective Institutions Project (EIP). Launched in early 2026, the site synthesizes five years of AI safety philanthropic advisory into a comprehensive guidebook aimed at persuading major donors to fund AI safety work.

The Foundation Layer Fund, managed by Tyler John, has facilitated over 100 grants exceeding $70 million across five intervention areas: alignment science, nonproliferation of dangerous capabilities, defensive technology, power distribution, and talent mobilization. The fund is available through Every.org.

The site argues that "philanthropists became the glue that held the world together" during the Cold War and can play a similar role in addressing AI risks. It appeals for diverse participation across political, geographic, and ideological lines, stating "we need all hands on deck."

History

Tyler M. John holds a PhD in philosophy from Rutgers University and was a Global Priorities Fellow at the Forethought Foundation for Global Priorities Research. He previously built Longview PhilanthropyOrganizationLongview PhilanthropyLongview Philanthropy is a philanthropic advisory organization founded in 2018 that has directed \$140M+ to longtermist causes (\$89M+ to AI risk), primarily through UHNW donor advising and managed...Quality: 45/100's AI advisory team and co-edited The Long View: Essays on Policy, Philanthropy, and the Long-term Future with Natalie Cargill (Longview's founder). He later founded the Effective Institutions Project (EIP), which specializes in AI safety, geopolitics, and power concentration.

EIP organized what Tyler John describes as "the first major town hall style meetings with Schmidt's P150 in 2023" with over 60 funders attending. The Foundation Layer site represents the culmination of this advisory work, consolidating recommendations into a publicly accessible guide.

Five-Pillar Strategy

The Foundation Layer proposes a five-part philanthropic strategy for addressing AI risk:

Pillar 1: Alignment Science

Ensuring AI systems reliably follow human instructions. Key focus areas include:

• Mechanistic interpretability: Understanding how AI models think by identifying and modifying their internal representations. Recent breakthroughs include Sparse Autoencoders that can identify and modify specific concepts within models.
• AI Control: Developed by Redwood ResearchOrganizationRedwood ResearchA nonprofit AI safety and security research organization founded in 2021, known for pioneering AI Control research, developing causal scrubbing interpretability methods, and conducting landmark ali...Quality: 78/100, this approach uses multiple AI systems to supervise and control potentially misaligned models.
• Additional research needs: resisting jailbreaking, preventing adversarial fine-tuning, ensuring faithful chain-of-thought reasoning.

The guide highlights that the UK AI Security Institute identified $50 million in valuable research projects but only had $10 million available, with 832 applications requesting a total of $320 million.

Pillar 2: Nonproliferation of Dangerous Capabilities

Three approaches to preventing dangerous AI capabilities from spreading:

1. Model Evaluations: Organizations like METROrganizationMETRMETR conducts pre-deployment dangerous capability evaluations for frontier AI labs (OpenAI, Anthropic, Google DeepMind), testing autonomous replication, cybersecurity, CBRN, and manipulation capabi...Quality: 66/100, Apollo ResearchOrganizationApollo ResearchApollo Research demonstrated in December 2024 that all six tested frontier models (including o1, Claude 3.5 Sonnet, Gemini 1.5 Pro) engage in scheming behaviors, with o1 maintaining deception in ov...Quality: 58/100, RAND Corporation, and AVERI systematically test AI systems for dangerous capabilities before deployment. The UK AI Safety Institute (AISI) conducts pre-deployment testing through formal agreements with major labs.
2. AI Company Security: Current AI company security is described as comparable to "normal start-up level." Model weights fit on a 5TB hard drive, making theft feasible.
3. Compute Governance: Leveraging the extreme concentration of AI chip supply (Nvidia designs, TSMC manufactures, ASML makes the lithography lasers) for tracking, verification, and compliance mechanisms.

Pillar 3: Defensive Technology

Building societal resilience against AI-enabled harms, with biodefense highlighted as the most neglected area:

• Far-UVC light can eliminate 90% of coronaviruses in 8 minutes
• Glycol vapors can achieve massive reductions in airborne pathogens within an hour
• Blueprint BiosecurityOrganizationBlueprint BiosecurityAn EA-funded biosecurity nonprofit founded in 2023 by Jake Swett, dedicated to achieving breakthroughs in pandemic prevention through far-UVC germicidal light, next-generation PPE, and glycol vapor...Quality: 60/100 is positioned to deploy tens of millions toward these solutions

Pillar 4: Distributing Power

Ensuring AGI benefits are broadly shared:

• AI for social coordination: Google DeepMind's "Habermas machine" AI mediator helped British groups find common ground on divisive issues
• Auditing AI decision-making: Requiring companies to publish and be audited against model specifications
• Humanity in the loop: Proof-of-human-personhood systems and standards for when AI can substitute for human judgment
• Economy after AGI: Research on post-AGI economics; the guide notes that only a handful of economists (Anton Korinek, Erik Brynjolfsson, Chad Jones, Daron Acemoglu, Philip Trammell) study these impacts

Pillar 5: Talent and Infrastructure

Mobilizing talent across sectors:

• Nonprofit organizations: METROrganizationMETRMETR conducts pre-deployment dangerous capability evaluations for frontier AI labs (OpenAI, Anthropic, Google DeepMind), testing autonomous replication, cybersecurity, CBRN, and manipulation capabi...Quality: 66/100, Transluce, Goodfire, and Apollo ResearchOrganizationApollo ResearchApollo Research demonstrated in December 2024 that all six tested frontier models (including o1, Claude 3.5 Sonnet, Gemini 1.5 Pro) engage in scheming behaviors, with o1 maintaining deception in ov...Quality: 58/100 for capability evaluation and model internals research
• Talent pipelines: BlueDot Impact identified 1,000+ new AI safety/governance roles; Impact Academy built a million-person database of global technical researchers
• State capacity: Horizon Institute for Public Service, Foundation for American Innovation, and TechCongress build government technical capacity
• Information sharing: Civic AI Security Project and California's SB 53 for public disclosure requirements

Funding Landscape

Recommended Funds

Recommended Advisors

Investment and Political Giving

The guide also covers investment vehicles (Juniper Ventures, Halcyon Ventures, Safe AI Fund, Entrepreneurs First, Seldon Lab) and political giving advisors (AI Policy Network, Americans for Responsible Innovation, Center for AI Safety Action Fund, Public First).

Key Claims and Statistics

Criticism

Limitations

• Single-author perspective: The guide represents the views of one advisor, not an institutional publication or peer-reviewed analysis.
• Not yet indexed or peer-reviewed: As of February 2026, the site is not indexed by search engines and has no third-party reviews or commentary.
• Verification challenges: Many statistics cited lack primary source links, making independent verification difficult.
• Potential conflicts of interest: Tyler John manages The Foundation Layer Fund and advises at EIP, which could create incentives to frame the funding landscape in ways that favor his advisory services.

Key Uncertainties

• Whether The Foundation Layer Fund's $70M+ figure represents grants facilitated, recommended, or directly disbursed
• The degree of overlap between the Foundation Layer Fund and Longview Philanthropy's grantmaking (given Tyler John's prior role at Longview)
• Whether the guide's AGI timeline estimates and risk framing are representative of expert consensus or reflect a particular perspective within the AI safety community