Foresight Institute

Theory of Change

Foresight Institute's theory of change centers on "differential technology development" -- the idea that safety and security-enhancing technologies should be accelerated while dangerous ones are slowed. CEO Allison Duettmann: "What we can bring to bear in general to the different technological risk paradigms is a very deep technical incentive of a community that cares a ton about this and is really good in creating security-first technologies."

The org identifies two primary AI safety strategies:

Computer security and cryptography approaches to AI alignment. Drawn from Mark Miller's object-capability security paradigm and Drexler/Miller's "Agoric Open Systems" papers (1988). The idea is that secure multi-agent architectures, formal verification, and privacy-preserving computation can provide safety guarantees that traditional alignment approaches cannot. Allison: "Our comparative advantage lies in bringing a computer security inspired lens to AI development."
Whole Brain Emulation (WBE). A $20M endowed Grand Prize for the first human brain emulation, plus annual fast grants. The theory is that WBE could be "a differential technology that could help with AI safety" -- producing AI systems based on human cognition that might be more interpretable and more aligned than pure ML systems. Allison notes this is pursued "maybe also out of sheer desperation" as AGI timelines shorten.

The broader intellectual framework aligns with Vitalik Buterin's d/acc (decentralized, democratic, differential defensive acceleration): build technologies that shift the offense/defense balance toward defense without concentrating power in centralized authorities.

What They Do

AI for Science & Safety Nodes (launching April 2026): Physical hubs in San Francisco and Berlin offering grant funding (~$3M/year total, typically $10K-$100K per grant), office space, and compute. Seven focus areas spanning AI security, privacy, cooperative AI, epistemics, neurotech/WBE, longevity biotech, and molecular nanotech. Named funders: Protocol Labs, Gigafund, 100 Plus Capital. ~40+ named grantees including CAIS (Dan Hendrycks), UC Berkeley (Dawn Song), Apart Research, OpenMined, Metaculus, MATS, and Cooperative AI Foundation.

Grants track record: In the first year after launching the grant program (August 2023), funded 10 projects totaling $440K. The 2023 990 filing shows $856,909 in total grants. The org claims "$4.5-5.5M in annual funding" but this figure significantly exceeds what 990 data confirms. Possible explanations include multi-year pledges, non-cash support (compute, office space), or aspirational figures.

Fellowship: One-year program, ~6% acceptance rate. Provides networking, travel to workshops, seminar access, and career counseling. Does not currently provide funding to fellows (a Fellowship Endowment is being raised to change this).

Prizes: Feynman Prize in Nanotechnology (since 1993, $5K/category annually, $250K Grand Prize unclaimed). Two prior winners later won Nobel Prizes -- David Baker (2024 Chemistry Nobel) and Sir Fraser Stoddart (2016 Chemistry Nobel). Norm Hardy Prize for usable security ($10K/year, since 2023). 2025 winners' work directly influenced the U.S. Cyber Trust Mark policy.

Events: Vision Weekend conferences (3 per year, $5K-$20K sponsor tiers), technical workshops (~50-80 curated participants per workshop on topics like cryptography/AI security), and monthly virtual seminars (150-400 members per group, 100K+ YouTube views).

Standout grantee success: Eon Systems (Foresight grantee) achieved fly brain emulation -- 139,255 neurons, ~50 million synaptic connections -- published in Nature (October 2024). The fly responds to sensory stimuli and exhibits biologically accurate behaviors.

Key People

Allison Duettmann (CEO/President since ~2020). MS Philosophy & Public Policy from LSE with AI Safety focus. Read the LW Sequences and MIRI publications while at LSE's EA community, then cold-emailed Foresight from Germany. Has presided over a 22x revenue increase since 2018. Co-authored "Gaming the Future" with Mark Miller and Christine Peterson.

Mark S. Miller (Senior Research Fellow). Pioneer of agoric computing and smart contracts. Designer of the E programming language. Google Research alumnus. Central intellectual figure for Foresight's security-first AI safety approach. His object-capability security research is a genuine and under-explored approach to AI safety.

Christine Peterson (Co-Founder, Senior Fellow). Coined "open source software" (1998). MIT alumna. On MIRI advisory board. Provides institutional memory and legitimacy from Foresight's 1986 founding; appears to have stepped back from operational leadership.

Team size is claimed at ~52 employees across 5 continents, but total salaries + wages of $887K (2024) suggest most are part-time, fellows, or contractors.

Money and Incentives

Revenue: $136K (2010) to $9.4M (2024) -- 69x growth in 14 years, almost entirely under Allison's leadership. 2024 saw a near-tripling from $3.6M (2023). Revenue is 85.6% contributions/donations, 6.2% program services (event fees), 7.9% sales of assets.

Grants vs. claims: The single most concerning financial finding is the gap between claimed and actual grantmaking. Foresight claims "$4.5-5.5M in annual funding." The AI Nodes page says ~$3M/year. But the 2023 990 shows only $856,909 in grants given. Even with 2024's $9.4M revenue, total expenses appear to be ~$3.8M. The most charitable explanation is that grants are ramping up with the 2026 AI Nodes launch, but the discrepancy warrants scrutiny.

Asset accumulation: Total assets grew from $3.1M to $8.9M in one year (2023-2024), meaning ~$5.8M was accumulated rather than deployed. The org simultaneously claims to be "funding-constrained" for grants.

Known funders: Protocol Labs, Gigafund, 100 Plus Capital (Sonia Arrison's fund, creating a board-funder conflict), FLI ($290K in 2023). Zero grants from Coefficient Giving/Open Philanthropy -- Foresight operates entirely outside the EA funding ecosystem. Donor identity for ~$8M in 2024 contributions is mostly opaque. Crypto donations accepted (BTC, ETH, SOL).

Compensation: Allison Duettmann $276K (2024), Beatrice Erkers $161K, Sherry Hull $100K. Total executive comp $548K = 14.6% of expenses. Reasonable for San Francisco.

Independence from AI labs: Zero compute credits, research contracts, or revenue streams from frontier labs. This is a genuine advantage for credibility -- Foresight has no financial incentives that could warp its safety advocacy.

What Others Say

LessWrong "grifty" critique: A poster described having their "grift sensors" triggered after being quickly accepted for a CEO interview and then invited to a $350 workshop. Worst case: "Foresight has identified the rationalist community as a group that's particularly gullible and bad at coordination, and were there to extract value." Foresight's reported defense: "small frugal nonprofit" with "oversubscribed programs." The defense was more credible at $502K revenue (2020) than at $9.4M (2024).

80K Hours on WBE: Rates WBE as "sometimes recommended" with significant caveats. Key concern: WBE neuroscience research "could be used to develop less safe types of AI" including neuromorphic AI. Bostrom (2014) argued neuromorphic AI is less safe than other forms. "We would have had to deal with two dangerous transitions" -- WBE arrival AND conventional AI advancement.

Vitalik Buterin on d/acc timeline risk: Acknowledges the strongest argument against d/acc: "if we have three year timelines until AGI, and another three years until superintelligence... we can't just accelerate the good, we also have to slow down the bad." Foresight's entire approach assumes enough time to build defensive infrastructure.

CAIS skepticism: The s-risks.org reviewer of Drexler's CAIS framework: "I'm fairly sceptical about these claims" regarding whether humans can retain strategic control over AI systems.

Positive signals: FLI co-funds events and gave a $290K grant. Anthropic researchers participate in Foresight workshops (Jason Clinton, Keri Warr, Dustin Li). Feynman Prize alumni include 2 Nobel laureates. MATS, CAIS, Apart Research, and Metaculus accept Foresight grants. A Manifund project exists to increase Foresight's AI safety funding.

What's Absent

No published conflict of interest policy despite clear board-funder overlaps (Sonia Arrison / 100 Plus Capital, Dean Tribble / Agoric).
No grant outcome reports. We know who received grants but not what they produced (except Eon Systems' fly brain Nature publication).
Major donor identity unknown for ~$8M in 2024 contributions.
No independent evaluation by any charity assessor or external reviewer.
No published annual report beyond mandatory 990 filings and an occasional LessWrong post.
Grant decision-making process is opaque: unnamed in-house reviewers and unnamed technical advisors; no published grant committee or recusal procedures.
$20M WBE Grand Prize endowment status unclear -- presented as a fundraising target, not as funded.
No formal collaboration with alignment-specific organizations (ARC, MIRI, Redwood) beyond Christine Peterson sitting on MIRI's advisory board.
Drexler's current role is undefined despite being co-founder and author of the CAIS framework that underpins Foresight's AI safety thinking.