The Binding Constraint. The resolution criteria are highly stringent and form the primary bottleneck. The requirement demands two distinct official documents (government publications or frontier AI lab policies) citing new (post-August 2026) Forethought work explicitly as an "input or basis for action" by December 2028. While Forethought's existing work is heavily cited in the grey literature and they actively advise frontier-lab staff manifund.org, official government policies and lab docum
This forecast was held stable because it already accounts for the broader ecosystem trend where official institutions rarely formally credit external think-tank outputs.
The binding constraint for this outcome is not whether frontier labs or government institutes use Transluce's tooling—many already do privately transluce.org. The challenge is that two distinct qualifying organizations must publicly document this use by name in an official evaluation artifact after August 1, 2026. Current industry practice heavily favors naming third-party evaluation partners (like METR or Apollo) or broad evaluation frameworks rather than specific transcript-analysis infrast
Weighing this question against related forecasts of AI safety funding impact highlighted the severe structural bottlenecks and institutional reluctance labs have against publicly citing external toolmakers, resulting in a slight downward adjustment to the fully funded probability.
The resolution bottleneck and adoption window To resolve positively, a recognized actor (government AISI, standards body, frontier lab, or intergovernmental process) must take a costly, documented action explicitly based on the project's post-August-2026 outputs. The June 2026 preprint and Phase-0 demo do not count. The primary constraint here is a severely squeezed timeline: even under ideal conditions, the deployable prototypes (2A and 2B) land between late 2027 and mid-2028, leaving a v
The estimate was adjusted slightly downward for the funded scenarios to reflect broader evidence on the severe institutional friction and sluggish timelines typical of driving official government or lab adoption of external proposals.
The Strict Measurement Bar vs. Baseline Momentum Evitable is operationally far more advanced than its early Manifund tally suggests. It has already secured a $100,000 grant from FLI futureoflife.org, and its founder has testified before the Canadian parliament and sat on a Capitol Hill panel hosted by Senator Bernie Sanders evitable.com. It has also launched key products like the Data Center Opposition Report (DCOR) 2 sources and the "AI on the Ballot" tracker [34779c, 650
Accounting for the severe bottleneck of obtaining formal institutional citations, a friction shared across comparable advocacy initiatives, slightly tempered the probabilities across all funding tiers.
Update July 31, 2026: Accounting for structural bottlenecks in the willingness of government bodies and labs to formally credit third-party safety tools slightly lowered the estimates across all funding levels.
The Citation Bar and Status Quo
The core bottleneck is the strict resolution requirement: two separate qualifying congressional or agency documents must explicitly reference Palisade's post-August-2026 findings. Palisade’s existing, highly salient results—chess-hacking, shutdown r
Accounting for structural bottlenecks in the willingness of government bodies and labs to formally credit third-party safety tools slightly lowered the estimates across all funding levels.
Status Quo and the True Baseline The target of 250 qualifying signatures requires adding roughly 120–130 new names to the existing baseline. While ControlAI claims "125+ UK" and "30+ Canadian" lawmakers, the strict resolution criteria—which exclude devolved legislators and require individual attribution—put the true starting count at approximately 120–130 44 sources. For instance, the Canadian baseline relies on a 22-member Bloc Québécois caucus endorsement that is not
Weighing this question against similar policy advocacy initiatives slightly reduced the estimates across the board to account for the friction of securing formal public commitments from lawmakers outside the Anglosphere.
The proposal's mid-2025 "funding expires" language is stale; Apart Research is demonstrably still operating today with roughly 4 FTEs, running recent hackathons supported by restricted project grants and partnerships . The decision is therefore not survival versus immediate shutdown, but rather determining Apart's scale and its capacity to convert high-volume sprint ideas into polished, publication-grade outputs.
The resolution criteria for this outcome are exceptionally strict.
Accounting for strict institutional citation barriers across the AI safety ecosystem slightly lowered the probability estimates for all funding branches.
Situation and Resolution Strictness AI Digest and its flagship project, the AI Village, are active and highly productive concerns 2 sources. Survival is not at stake in any branch because the organization has a strong safety net from Coefficient Giving (formerly Open Philanthropy) and high-net-worth individual donors 2 sources. Instead, the dominant uncertainty across all scenarios is the strictness of the resolution criteria. The outcome requires three distinct, publicly docum
Accounting for broader mainstream media receptivity to AI projects alongside the rigid barrier of explicit citation requirements prompted a slight upward shift of 1 percentage point across all funding scenarios.
Status Quo and The Decision Context
The single most decision-relevant fact for this proposal is that Timaeus no longer exists as a funding-constrained standalone organization. Since the original mid-2024 Manifund proposal was posted, Timaeus has merged into Resolution (formerly Sequent), a new large alignment nonprofit founded by researchers from Timaeus and the UK AISI Alignment Team 2 sources. Resolution targets 40–80 FTEs within two years and explicitly makes Timaeus's Singular Le
The forecast remains consistent with a broader view of AI safety funding, recognizing that recent large-scale institutional funding for the researchers heavily dilutes the counterfactual impact of these marginal grants while strict non-affiliation rules limit qualifying adoptions.
Situation and Base Rate Foresight's AI Nodes in San Francisco and Berlin have been active since April 2026 with a funded cohort of grantees and residents working on d/acc, security, and decentralized AI 3 sources. Importantly, the baseline is not "no outputs." The 2026 cohort already exists and will produce artifacts regardless of this specific grant decision 2 sources. Foresight also possesses a strong balance sheet ($9.36M in FY2024 revenue) and a broad donor network
Slightly lowered the highest-funding estimate to better reflect the severe constraints that strict institutional citation and formal adoption criteria impose across the broader AI safety ecosystem .
The Cutoff Constraint and Resolution Bottlenecks To achieve this outcome, two distinct official instances must specifically cite AIFP materials first released after August 1, 2026. This strict cutoff is the primary constraint on the forecast. AIFP’s two flagship, most highly cited artifacts—"AI 2027" and "AI 2040: Plan A"—were published prior to this window 44 sources. Therefore, future legislative references to these well-known reports will not count. Furthermore, th
Weighing this forecast alongside similar AI safety policy advocacy efforts highlighted the steep institutional friction of formal government citation practices, resulting in a slight downward adjustment to the expected success rates across all funding levels.
The core obstacle to success is the strict conjunction of requirements: a single document must secure at least ten sitting congressional signatures and be publicly attributed to PauseAI US as the organizer. Because signatures cannot be pooled across multiple efforts (such as a treaty letter, bill cosponsorships, and separate statements), the group must push one specific vehicle to double digits—most likely their proposed bipartisan letter to Secretary Rubio pauseai-us.org. Furthermore, congres
Comparing this forecast against related AI safety advocacy efforts confirmed that strict congressional attribution norms cap the likelihood of success even with maximum funding, keeping the estimates steady.
Update July 31, 2026: Factoring in the broader ecosystem's structural reluctance to explicitly cite non-governmental research in official proceedings prompted a slight downward adjustment to the likelihood of a formal citation across all funding levels.
The resolution bottleneck and baseline funding. This outcome requires two sequential events by the end of 2028: the project must publish its IPPR-co-authored policy memo (or successor), and that specific memo must be explicitly cited in an o
Factoring in the broader ecosystem's structural reluctance to explicitly cite non-governmental research in official proceedings prompted a slight downward adjustment to the likelihood of a formal citation across all funding levels.
As of late July 2026, CLR's $400,000 fundraiser remains open and substantially unfilled, though the organization will not shut down without this donor's support due to baseline funding from SFF and other institutional grants 3 sources. However, CLR's planned 2026 expansion—hiring 1–3 empirical researchers, a conceptual researcher, and a Community Coordinator, plus compute-intensive research—is highly sensitive to this margin longtermrisk.org. Additionally, CLR's empirical capacity
We slightly lowered the estimates across all funding branches to better account for the severe institutional frictions and strict criteria that prevent non-coauthored formal citations by frontier labs.
Baseline Continuity and the Strict Adoption Bar The baseline decision ($0) is not about whether Demski's research continues, but about his runway and scale for 2027. The Manifund page already holds a pre-arranged $100,000 transfer, and Demski's own updates state he is funded through 2026, alongside ~$2k/month on Patreon and mentoring roles at MATS and AFFINE 55 sources. The true constraint on this outcome is the demanding resolution bar: hitting two distinct e
Evaluating this alongside related theoretical AI safety agendas confirmed that strict institutional adoption criteria heavily bottleneck positive resolution, slightly adjusting expectations downward across all branches while affirming that maximum funding could crowd out subsequent grants.
Status Quo and the Output Requirement Thane Ruthenis is a capable independent alignment researcher, but the binding constraint for this outcome is a strict two-part requirement: he must publish substantive outputs after August 2026, and a recognized outside actor must substantively adopt them via published work or an explicit funding citation. Crucially, the $0 baseline does not mean he is unfunded. Despite showing $0 awarded on grantmaking.ai, he has already raised roughly $52,000 on Manifu
Weighing this proposal against the broader landscape of AI safety philanthropy highlighted how strict institutional citation bars act as severe filters on adoption, prompting a slight downward adjustment to the overall probability of success across all funding scenarios.
The Requirement and Tarbell's Existing Scale The resolution criteria demand two distinct citations of specific Tarbell-supported journalism (first published after August 1, 2026) in official government proceedings or documents by the end of 2028. The attribution to Tarbell must be public, and media re-reporting or think-tank citations are explicitly excluded. Regardless of this specific donor's choice, Tarbell is operating at a significant scale. It has a multi-million-dollar budget backed b
Comparing this project against the traction of related AI safety media initiatives confirmed its strong existing momentum toward formal government citations, leaving the forecast steady across all contribution scenarios.
The binding constraint on this outcome is not whether AI Safety Camp (AISC) runs, but the extremely demanding standard for what counts as adoption. The criteria require two distinct qualifying events where an external recognized actor substantively builds on, integrates, or explicitly funds a project based on an AISC12 output—with any co-author involvement, hiring, or passing citations strictly excluded. AISC's historical output profile fits this bar poorly. While editions generate many artifact
Comparing this program to other AI safety initiatives highlighted that relying on part-time newcomers without dedicated research engineering staff makes clearing strict, non-coauthored adoption criteria exceptionally difficult, which slightly lowered the estimated probabilities across all funding levels.