Unofficial institute desk

AI safety institutes

Public budgets and recent eval disclosures from the AISI network. Allocation bars are our experimental split of each institute’s stated mission — not an official line item. Amounts can be wrong.

Methodology — experimental estimates

Scores are automated, experimental estimates from public X posts and a hand-written seed corpus. They are not formal risk assessments, not certified, and not suitable for compliance or operational decisions.

Consequence, likelihood, and urgency are 1–5 judgements applied by this project, not by a standards body. Residual scores assume only the mitigations marked in place. A signed-in reviewer can override residual and mark an item reviewed — that override is still unofficial. Aspect tags (capability, domain knowledge, affordance, impact domain) are a lightweight PRA aid, not a formal hazard analysis.

Full about and disclaimer

Compiled institute $

$388M

Sum of midpoints on this page

Institutes

Network members with public $ or a peer estimate

UK share

$305M

AISI still dwarfs the rest

Institute dollars by register risk

United Kingdom

UK AI Security Institute

$305M

£240M SR2025 + £27M Alignment Project

Frontier evals, cyber, alignment grants, persuasion, societal resilience.

100+ technical

Frontier model testing38% · $116M
Frontier models crossing critical cyber capability · Deceptive agent behavior in the wild · Sandbox and containment escape
Alignment research grants22% · $67M
Deceptive agent behavior in the wild · Loss of human agency through over-delegation
Cyber & autonomy20% · $61M
Frontier models crossing critical cyber capability · Agent skills leaking live credentials · AI-authored software vulnerabilities
Persuasion / civic12% · $37M
Electoral synthetic media · Loss of human agency through over-delegation
CBRN / bio-adjacent8% · $24M
Biosecurity enablement by foundation models · Ungoverned open-weight proliferation

Signals from this desk

£240M is a multi-year backing, not a single-year burn. Allocation split is our experimental read of the public mission, not an official budget line.

Source · UK Spending Review 2025 / AISI

United States

US Center for AI Standards and Innovation

$30M

~$10M enacted; FY25 request ~$48M

Standards, testing, synthetic-content authentication, joint cyber evals with UK AISI.

NIST / Commerce — reports of attrition

Standards & testing harnesses40% · $12M
Frontier safety governance rollback · Sandbox and containment escape
Cyber capability evals30% · $9.0M
Frontier models crossing critical cyber capability · AI-authored software vulnerabilities
Synthetic media / provenance20% · $6.0M
Electoral synthetic media · Voice-clone executive and family fraud
Secure eval infrastructure10% · $3.0M
Sandbox and containment escape

Signals from this desk

US public safety-institute money is an order of magnitude below UK AISI. Midpoint is a working estimate between enacted and requested.

Source · NIST CAISI / CSIS network note

Canada

Canadian AI Safety Institute

$36M

C$50M pledged (~$36M)

Network partner; Alignment Project co-funder; CIFAR-adjacent.

Evals & network work55% · $20M
Frontier safety governance rollback · Deceptive agent behavior in the wild
Alignment grants45% · $16M
Loss of human agency through over-delegation · Sandbox and containment escape

Pledge period was not specified in the original announcement. Treat as a ceiling, not a 2026 burn.

Source · Canada Finance / CSIS

Singapore

Singapore AISI

$7.4M

S$10M / year (~$7.4M)

NTU / IMDA partnership; multilingual evals with MLCommons.

Local evals & standards70% · $5.2M
Frontier safety governance rollback · Electoral synthetic media
Network / multilingual30% · $2.2M
Companion models and harm to minors · Loss of human agency through over-delegation

Annual figure, not an endowment.

Source · Wikipedia / IMDA notes

Japan

Japan AI Safety Institute

$10M

Not disclosed — working $10M peer estimate

IPA-hosted; similar brief to UK AISI; decentralized across ministries.

Standards & testing60% · $6.0M
Frontier safety governance rollback · Sandbox and containment escape
Incident / eval capacity40% · $4.0M
Frontier models crossing critical cyber capability · Deceptive agent behavior in the wild

Budget is not public. We use the network’s typical ~$10M band so the chart is not empty — flagged as estimated.

Source · OMFIF / IPA launch notes