← Back to Research

Decision Forecasts on SFF's 2025 AI-Safety Grantees

For each of thirty grantees: nothing in 2026, the 2025 amount again, or double it.

OrganizationDoubled 2026 allocation
0%Probability of uptake by end of 2028100%
DeltaWorld model
Oxford China Policy Lab
New work cited as a basis for action in 2 official government or lab policy documents
$1.44M
2038
+18
PIBBSS
New outputs used or built on by 2 unaffiliated safety actors
$2.03M
2437
+13−2
Palisade Research
New findings cited in 2 US congressional hearings or official government publications
$2.27M
2537
+12held
ACS Research
New outputs used or built on by 2 unaffiliated safety actors
$1.41M
4961
+12−2
CivAI
2 national legislators, bodies or civil-society organizations join or cite a new campaign
$1.20M
3648
+12
Metaculus
Forecasts or reports used or cited in 2 official government settings or lab policy documents
$1.50M
3140
+9
CLTR
New work cited as a basis for action in 2 official government or lab policy documents
$1.13M
4756
+9
Center for Humane Technology
2 national legislators, bodies or civil-society organizations join or cite a new campaign
$936k
5968
+9
AI Policy Institute
New work cited as a basis for action in 2 official government or lab policy documents
$3.27M
917
+8
CAIS Action Fund
New work cited as a basis for action in 2 official government or lab policy documents
$1.54M
1018
+8
Encode AI Corporation
2 national legislators, bodies or civil-society organizations join or cite a new campaign
$1.03M
7785
+8
The Future Society
New work cited as a basis for action in 2 official government or lab policy documents
$672k
1321
+8
SaferAI
New work cited as a basis for action in 2 official government or lab policy documents
$622k
2129
+8
Secure AI Project
New work cited as a basis for action in 2 official government or lab policy documents
$766k
2936
+7
AI Futures Project
New materials used or cited in 2 official government settings or lab policy documents
$4.07M
1420
+6
RAND (Technology and Security Policy Center)
New work cited as a basis for action in 2 official government or lab policy documents
$2.04M
4450
+6
Tarbell Center for AI Journalism
Supported journalism cited in 2 official government proceedings or publications
$1.57M
3743
+6
GovAI
New work cited as a basis for action in 2 official government or lab policy documents
$1.51M
3440
+6
Simon Institute for Longterm Governance
New work cited as a basis for action in 2 official government or lab policy documents
$800k
915
+6
Foundation for American Innovation
New work cited as a basis for action in 2 official government or lab policy documents
$1.55M
4449
+5
IAPS
New work cited as a basis for action in 2 official government or lab policy documents
$1.14M
3237
+5
Singapore AI Safety Hub
New work cited as a basis for action in 2 official government or lab policy documents
$672k
712
+5
MIRI
New outputs cited as a basis for action in 2 official policy documents or legislative records
$3.21M
812
+4held
FAR AI
New outputs used or built on by 2 unaffiliated safety actors
$1.84M
6771
+4−2
Oxford Martin AI Governance Initiative
New work cited as a basis for action in 2 official government or lab policy documents
$622k
2832
+4
CAIS
New outputs used or built on by 2 unaffiliated safety actors
$578k
7680
+4−2
MATS
2 scholar outputs built on by unaffiliated safety actors
$578k
7780
+3
Timaeus
Methods or tools used or extended by 2 unaffiliated safety actors
$552k
2528
+3held
METR
Evaluations documented as decision inputs by 2 frontier labs or safety institutes
$1.10M
8991
+2held
AI Lab Watch
New work cited as a basis for action in 2 official government or lab policy documents
$742k
24
+2
Bar: $0 (left) to double the 2025 amount (right); tick: the 2025 amount again. Delta in percentage points. Only work first produced after October 1, 2026 counts. Sorted by delta. World model: how far the consistency pass moved the doubled-allocation forecast, in percentage points — blank where it left the row alone, "held" where it reviewed the row and kept the number; hover for the other branches. Click a bar to open the run.