Coefficient Giving

📋Page Status

Page Type:ContentStyle Guide →Standard knowledge base article

Quality:55 (Adequate)⚠️

Importance:62 (Useful)

Last edited:2026-01-29 (3 days ago)

Words:3.6k

Structure:

📊 21📈 2🔗 0📚 50•10%Score: 13/15

LLM Summary:Coefficient Giving (formerly Open Philanthropy) has directed $4B+ in grants since 2014, including $336M to AI safety (~60% of external funding). The organization spent ~$50M on AI safety in 2024, with 68% going to evaluations/benchmarking, and launched a $40M Technical AI Safety RFP in 2025 covering 21 research areas with 2-week EOI response times.

Issues (2):

QualityRated 55 but structure suggests 87 (underrated by 32 points)
Links12 links could use <R> components

Quick Assessment

Dimension	Assessment	Evidence
Scale	Dominant	$4B+ total grants; ≈$46M AI safety in 2023
Structure	13 cause-specific funds	Multi-donor pooled funds since Nov 2025 rebrand
AI Safety Focus	Leading funder	$336M+ to AI safety since 2014; ≈60% of external AI safety funding
Application Model	Rolling RFPs + regranting	300-word EOI, 2-week response; supports platforms like Manifund
Transparency	High	Public grants database, annual progress reports
Key Funders	Good Ventures (primary)	Dustin Moskovitz & Cari Tuna; expanding to multi-donor model

Organization Details

Attribute	Details
Full Name	Coefficient Giving (formerly Open Philanthropy)
Type	Philanthropic Advising and Funding Organization
Legal Structure	LLC (independent since 2017)
Founded	2014 (as GiveWell outgrowth); 2017 (independent); 2025 (rebranded)
Total Grants	$4+ billion (as of June 2025)
AI Safety Grants	$336+ million (≈12% of total)
2024 AI Safety Spend	≈$50 million committed
Leadership	Alexander Berger (CEO), Holden Karnofsky (Board)
Location	San Francisco, California
Website	coefficientgiving.org
Grants Database	coefficientgiving.org/grants

Overview

Coefficient Giving is a major philanthropic organization that has directed over $4 billion in grants since 2014 across global health, AI safety, pandemic preparedness, farm animal welfare, and other cause areas. In November 2025, the organization rebranded from Open Philanthropy to Coefficient Giving, signaling an expansion from serving primarily one anchor donor (Good Ventures, the foundation of Dustin Moskovitz and Cari Tuna) to operating 13 cause-specific funds open to multiple philanthropists. The name “Coefficient” reflects the organization’s goal of multiplying impact through research, grantmaking, and partnerships—with “co” nodding to collaboration and “efficient” reflecting their unusual focus on cost-effectiveness.

Coefficient Giving is widely considered the largest funder of AI safety work globally. Since 2014, approximately $336 million (12% of total grants) has gone to AI safety research and governance, with roughly $46 million deployed in 2023 alone—making it the dominant external funder in a field where most safety research happens inside frontier AI labs. The organization’s Navigating Transformative AI Fund supports technical AI safety research, AI governance and policy work, and capacity building, with a $40 million Technical AI Safety RFP launched in 2025 covering 21 research areas.

The organization distinguishes itself through its strategic cause selection methodology—identifying problems that are large, tractable, and neglected relative to their size. This approach, combined with a willingness to fund speculative research and support multiple funding mechanisms (direct grants, regranting programs, pooled funds), has made Coefficient Giving central to the effective altruism funding ecosystem. However, critics have noted concerns about funding concentration, the slow pace of spending relative to the scale of AI risks, and heavy focus on evaluations over alignment research in recent technical AI safety grants.

History and Evolution

Origins (2011-2017)

Coefficient Giving traces its origins to 2011 when GiveWell, the charity evaluator founded by Holden Karnofsky and Elie Hassenfeld, began advising Good Ventures on how to deploy Dustin Moskovitz’s philanthropic capital effectively. Good Ventures was established by Moskovitz (Facebook co-founder, net worth ≈$12 billion) and Cari Tuna in 2011. By 2014, this advising relationship formalized into “Open Philanthropy” as a distinct project within GiveWell, focused on identifying high-impact giving opportunities across a broader range of cause areas than GiveWell’s traditional global health focus.

In 2017, Open Philanthropy spun off from GiveWell as an independent LLC, enabling it to pursue its own strategic priorities while GiveWell continued focusing on evidence-backed global health interventions. The separation reflected diverging methodologies: GiveWell prioritizes robust evidence of effectiveness, while Open Philanthropy embraced “hits-based giving”—funding speculative, high-variance projects where a few major successes could justify many failures.

Growth and AI Safety Focus (2015-2024)

Open Philanthropy began supporting AI safety work in 2015, when the field was nascent and institutional support was minimal. Early grants helped establish foundational organizations including the Machine Intelligence Research Institute (MIRI), the Center for Human-Compatible AI at UC Berkeley, and the Future of Humanity Institute at Oxford. By 2023, AI safety had become Open Philanthropy’s largest longtermist cause area, reflecting growing concern about advanced AI risks among the leadership team.

Year	AI Safety Milestone
2015	First AI safety grants; field had ≈10 full-time researchers
2017	Independent organization; Holden Karnofsky publishes AI concerns
2019	AI safety spending exceeds $20M annually
2022	$150M Regranting Challenge launched (not AI-specific)
2023	≈$46M AI safety spending; largest funder in the field
2024	≈$50M committed; 68% to evaluations/benchmarking
2025	Rebrand to Coefficient Giving; $40M Technical AI Safety RFP

The November 2025 Rebrand

On November 18, 2025, Open Philanthropy announced its rebranding to Coefficient Giving. The change reflected several strategic shifts:

Multi-Donor Expansion: The organization moved from primarily serving Good Ventures to operating pooled funds open to any philanthropist. In 2024, Coefficient directed over $100 million from donors besides Good Ventures; by 2025, non-Good Ventures funding had more than doubled.

Brand Clarity: The “Open Philanthropy” name created confusion—journalists mistook them for OpenAI, potential grantees confused them with Open Society Foundations. “Coefficient” provided a distinctive identity.

Structural Reorganization: The organization restructured from program areas to 13 distinct funds, each with dedicated leadership and transparent goals, allowing donors to support specific causes at scale.

Organizational Structure

Loading diagram...

The 13 Funds Model

Since the November 2025 rebrand, Coefficient Giving operates through 13 cause-specific funds, each pooling money from multiple donors:

Fund	Focus	Key Activities
Navigating Transformative AI	AI safety & governance	Technical research, policy, capacity building
Biosecurity & Pandemic Preparedness	Catastrophic bio risks	Research, policy, infrastructure
Global Catastrophic Risks Opportunities	Cross-cutting x-risk work	Ecosystem support, foundational work
Science and Global Health R&D	Neglected disease research	TB, malaria, high-risk transformational science
Global Health Policy	Policy for health impact	Lead exposure, air pollution
Global Aid Policy	Development effectiveness	Evidence-based aid policy
Farm Animal Welfare	Factory farming reform	Welfare reforms, alternative proteins
Effective Giving and Careers	EA movement building	Giving What We Can, 80,000 Hours
Abundance & Growth	Economic prosperity	$120M launched 2025 for scientific progress
Criminal Justice Reform	US criminal justice	Bail reform, prosecutorial accountability
Land Use Reform	Housing and development	YIMBY policy, zoning reform
Immigration Policy	Immigration reform	Policy research and advocacy
Other Global Health	Remaining health causes	Malaria, deworming, direct cash transfers

Navigating Transformative AI Fund

The Navigating Transformative AI Fund is Coefficient’s primary vehicle for AI-related grantmaking, supporting:

Technical AI Safety Research: Work aimed at making advanced AI systems more trustworthy, robust, controllable, and aligned. This includes interpretability research, robustness to adversarial inputs, scalable oversight methods, and understanding emergent capabilities.

AI Governance and Policy: Frameworks for safe, secure, and responsibly managed AI development, including export controls, compute governance, international coordination, and corporate governance mechanisms.

Capacity Building: Growing and strengthening the field of researchers and practitioners working on AI challenges, including training programs, career development, and institutional infrastructure.

Short-Timeline Projects: New projects expected to be particularly impactful if timelines to transformative AI are short, reflecting Coefficient’s view that advanced AI could emerge within the next 5-15 years.

Regrantor Model

Component	Description
Selection	OP identifies trusted individuals with relevant expertise
Budget	Each regrantor receives $200K - $2M to distribute
Autonomy	Regrantors make independent decisions within guidelines
Reporting	Regrantors document grants, OP maintains oversight
Renewal	Strong performers may receive additional budgets

Regrantor Criteria

Criterion	Description
Domain Expertise	Deep knowledge in cause area
Community Connections	Know who does good work
Judgment	Track record of good decisions
Capacity	Time to evaluate and make grants
Values Alignment	Share EA/longtermist priorities

AI Safety Grantmaking

Major AI Safety Grantees (2024)

Coefficient’s largest 2024 AI safety grants reflect priorities across evaluations, interpretability, and theoretical alignment work:

Grantee	Amount	Focus	Notes
Center for AI Safety	$8.5M	Field building, research	Training programs, compute grants, advocacy
Redwood Research	$6.2M	Alignment research	Interpretability, control research; $21M+ total from OP
MIRI	$4.1M	Theoretical alignment	Agent foundations, deceptive alignment
Epoch AI	≈$3M	AI forecasting	Compute trends, capability timelines
METR (formerly ARC Evals)	≈$3M	Capability evaluations	Model evaluations used by labs and governments
AI Safety Camp	≈$500K	Talent pipeline	Intensive research programs
Various Individuals	≈$10M	Researchers, fellowships	PhDs, postdocs, independent researchers

2024 Technical AI Safety Funding Breakdown

An analysis of Open Philanthropy’s Technical AI Safety funding revealed the following distribution of the $28M recorded in their database:

Research Area	Percentage	Amount (~)	Assessment
Evaluations/Benchmarking	68%	$19M	Primary focus; critics note AI Safety Institutes already well-resourced
Interpretability	≈10%	≈$3M	Mechanistic interpretability, circuit analysis
Robustness	≈5%	≈$1.5M	Adversarial robustness, red-teaming
Value Alignment	≈5%	≈$1.5M	RLHF alternatives, preference learning
Field Building	≈5%	≈$1.5M	Training programs, community
Forecasting	≈3%	≈$1M	Timelines, capabilities
Other	≈4%	≈$1M	Governance research, miscellaneous

Note: The $28M figure underestimates total 2024 spending as some approved grants had not been posted to the database at time of analysis. Coefficient acknowledged spending “roughly $50 million” on technical AI safety in 2024.

Historical Major AI Safety Grants

Grantee	Total (All Years)	Period	Notable Impact
MIRI	$14M+	2014-2024	Agent foundations, embedded agency
Redwood Research	$21M+	2021-2024	Interpretability methods, control research
Center for AI Safety	$15M+	2022-2024	Compute cluster, training programs
Future of Humanity Institute	$10M+	2015-2024	Strategic analysis (closed 2024)
Center for Human-Compatible AI	$8M+	2016-2024	Stuart Russell’s CHAI lab
Anthropic	$0 directly	N/A	VC-funded; OP staff invested personally
Long-Term Future Fund	$3.15M	2019-2024	Regranting to LTFF for distribution

2025 Technical AI Safety RFP

In early 2025, Coefficient launched a $40 million Request for Proposals across 21 research areas, with funding available for substantially more based on application quality. Key features:

Priority Research Areas (starred items are especially prioritized):

Category	Research Areas
Alignment Foundations	Alternatives to adversarial training, alignment faking, scalable oversight*
Interpretability	Mechanistic interpretability*, representation engineering, probing
Evaluation	Dangerous capability evaluations, propensity evaluations, automated red-teaming
Robustness	Adversarial robustness, distribution shift, specification gaming
Governance-Adjacent	AI governance research, responsible scaling policies

Grant Characteristics:

Aspect	Details
Size Range	API credits ($1-10K) to seed funding for new orgs ($1M+)
Application	300-word expression of interest (EOI)
Response Time	Within 2 weeks of EOI submission
Decision Timeline	4-8 weeks for full proposals
Eligibility	Academic researchers, nonprofits, independent researchers, new orgs

Regranting Ecosystem

Coefficient Giving supports multiple regranting platforms and mechanisms to achieve faster, more distributed funding decisions. This represents a deliberate strategy to complement slower direct grantmaking with nimble, expert-driven allocation.

Funding Flow Through Regranting

Loading diagram...

Long-Term Future Fund (LTFF)

The Long-Term Future Fund is a committee-based grantmaking fund that receives significant support from Coefficient. About half of LTFF funding historically comes from Open Philanthropy donations.

Aspect	Details
Annual Volume	≈$6.7M (2023)
AI Safety Portion	≈$4.3M (≈65% of grants)
Grant Count	≈200 grants per year
Median Grant	≈$15-30K
Decision Model	Committee of fund managers
Transparency	High (public grant reports)

LTFF grants tend toward smaller, faster decisions than direct Coefficient grants, serving researchers and projects that may not yet warrant Coefficient’s full evaluation process.

Manifund AI Safety Regranting

Manifund operates a distinct regranting model where individual experts receive budgets to make independent funding decisions. For 2025, Manifund raised $2.25 million and announced their first 10 regrantors.

Named 2025 Regrantors:

Regrantor	Budget	Background	Focus
Evan Hubinger	$450K	Anthropic AGI Safety Researcher, former LTFF manager	Technical AI safety
Ryan Kidd	≈$100K+	Co-director of SERI MATS	Emerging talent
Marius Hobbhahn	≈$100K+	CEO of Apollo Research	Evaluations, scheming
Lisa Thiergart	≈$100K+	Director at SL5 Task Force, former MIRI	Governance
Gavin Leech	≈$100K+	Cofounder Arb Research	Research reviews
Dan Hendrycks	≈$100K+	Director of CAIS	Safety research
Adam Gleave	≈$100K+	CEO of FAR AI	Adversarial robustness

Manifund Regranting Characteristics:

Feature	Details
Speed	Grant to bank account in under 1 week
Typical Grant Size	$5K-$50K
Decision Authority	Solo regrantor decisions
Oversight	Manifund reviews but doesn’t approve
Risk Tolerance	High (encourages speculative grants)

Notable Manifund Grants:

Project	Amount	Regrantors	Impact
Timaeus (DevInterp)	$143,200	Evan Hubinger, Rachel Weinberg, Marcus Abramovitch, Ryan Kidd	First funding; accelerated research months
ChinaTalk	$37,000	Joel Becker, Evan Hubinger	Coverage of China/AI, including DeepSeek
Shallow Review 2024	$9,000	Neel Nanda, Ryan Kidd	Induced further $5K from OpenPhil

Survival and Flourishing Fund (SFF)

The Survival and Flourishing Fund uses a unique “S-process” algorithm for grant allocation, primarily funded by Jaan Tallinn (Skype co-founder). While Coefficient and SFF are independent, they share many grantees and strategic priorities.

Aspect	Coefficient	SFF
2024 Volume	≈$650M total	≈$24M
AI Safety %	≈12%	≈86% ($20M)
Decision Model	Staff + regrantors	S-process algorithm
Speed	Rolling	Twice yearly rounds
Overlap	High	High

How to Apply for Funding

Direct Application to Coefficient

The most straightforward path for substantial funding requests:

Step	Details	Timeline
1. Check RFPs	Review active Requests for Proposals	Ongoing
2. Submit EOI	300-word expression of interest describing project	N/A
3. Initial Response	Coefficient responds with interest level	2 weeks
4. Full Proposal	If invited, submit detailed proposal with budget	2-4 weeks to prepare
5. Due Diligence	Coefficient evaluates organization and proposal	4-8 weeks
6. Decision	Grant approval or rejection	Total: 2-4 months

Tips for Applicants (from Coefficient’s guidance):

The bar is intentionally low for submitting expressions of interest. Key failure modes to avoid include not demonstrating understanding of prior work (read papers linked in relevant RFP sections) and not demonstrating that your team has prior experience with ML projects. Even uncertain proposals are worth submitting as the RFP is partly an experiment to understand funding demand.

Via Regranting Platforms

Faster and more accessible for smaller grants:

Platform	Best For	How to Apply
Manifund	$5-50K projects, emerging researchers	Create project on manifund.org, contact regrantors directly
LTFF	$10-100K, established track record	Apply via EA Funds
SFF	$100K+, established organizations	Apply during S-process rounds

Finding Regrantors

Many regrantors are reachable through:

Direct outreach: Email or social media (many are publicly active on Twitter/X, LessWrong)
EA communities: EA Forum, Alignment Forum, local EA groups
Professional networks: AI safety conferences (NeurIPS safety track, ICML), SERI MATS alumni
Manifund platform: Create project and regrantors may proactively reach out

Comparison with Other AI Safety Funders

Aspect	Coefficient Giving	LTFF	SFF	Manifund
2024 AI Safety Volume	≈$50M	≈$4.3M	≈$20M	≈$2M
Total Assets	Good Ventures ($12B+)	Pool of donors	Jaan Tallinn	Donors
Decision Model	Staff + regrantors	Committee	S-process algorithm	Individual regrantors
Typical Grant Size	$100K-$5M	$15-100K	$100K-$2M	$5-50K
Speed (EOI to decision)	2-4 months	1-3 months	6 months (rounds)	Under 2 weeks
Transparency	Medium (public database)	High (detailed reports)	High (S-process public)	Very high (live on platform)
Risk Tolerance	Medium	Medium-High	Medium	High
Best For	Major grants, established orgs	Growing researchers	Established orgs	Early-stage, speculative

Funding Gap Analysis

According to an overview of AI safety funding, total external philanthropic AI safety funding (≈$100M annually) is dwarfed by:

Comparison	Amount	Ratio to Safety Funding
Generative AI Investment (2023)	≈$24B	240:1
Frontier Lab Safety Budgets	≈$500M+ combined	5:1
US Government AI R&D	≈$3B annually	30:1

This funding gap is a persistent concern in the AI safety community, though Coefficient and other funders argue that talent constraints, not funding, are often the binding limitation.

Critical Assessment

Strengths

Scale and Stability: With Good Ventures’ multi-billion dollar backing, Coefficient can make commitments that smaller funders cannot. This enables multi-year organizational support, compute grants, and substantial research programs.

Strategic Sophistication: The organization’s cause selection methodology and research depth (public writeups, shallow investigations, deep dives) provides unusually transparent reasoning for grant decisions.

Ecosystem Building: By funding LTFF, Manifund, and other regranting mechanisms, Coefficient amplifies its reach while maintaining quality through trusted intermediaries.

Hits-Based Giving: Willingness to fund speculative research acknowledges that transformative progress often comes from unexpected directions, though this increases variance in outcomes.

Limitations and Criticisms

Funding Concentration: With Coefficient representing ~60% of external AI safety funding, the field is heavily dependent on one organization’s worldview and priorities. Critics note this could lead to “possible solutions being overlooked or assumptions no longer being questioned.”

Evaluation Focus: The heavy focus on evaluations/benchmarking (68% of 2024 technical grants) has drawn criticism. As one researcher noted, “This looks much worse than I thought it would, both in terms of funding underdeployment, and in terms of overfocusing on evals.” Critics argue AI Safety Institutes are already well-resourced for evaluation work.

Alignment Neglect: Some researchers express disappointment that “there’s so little emphasis in this RFP about alignment, i.e. research on how to build an AI system that is doing what its developer intended it to do.”

Slow Spending: Coefficient has acknowledged that “in retrospect, our rate of spending was too slow, and we should have been more aggressively expanding support for technical AI safety work earlier.” Key reasons cited include difficulty making qualified senior hires and disappointment with returns to past spending.

Grants Database Limitations: The public grants database “offers an increasingly inaccurate picture” of Coefficient’s work, as it generally excludes funding advised from non-Good Ventures donors. Coefficient is considering deprecating it.

Strategic Questions

Question	Context
Funding deployment rate	Is $50M/year appropriate given AI development pace?
Evaluation vs alignment balance	Should more funding go to core alignment research?
Lab relationships	How to maintain independence while funding lab-adjacent work?
Multi-donor model	Will expanding beyond Good Ventures change priorities?
Talent vs funding constraint	Is the field truly talent-constrained, or is this justifying underspending?

Sources and Citations

Primary Sources