RAND Corporation
Canary joint project with METR
Scorecard
Think tank with CBRN and security expertise; AEF founding member.
Nonprofit, Santa Monica, US. Independent referee; listed with independent referees. Confidence med. Domains: bio, cyber, assurance.
No floor triggered 6/8 evidenced · Retrieved & confirmed Scores by preset: Lab procurement –; Regulator or auditor –; Public trust –; Equal weights –. Weakest evidenced dimension: governance 2/4. No evidenced dimension below 2 — not a pass. No floor triggered means no disqualifying conflict is on file, not a pass. Partial evidence (6/8).
The weighted numbers are hidden until you ask for them: two readers with different priorities see different numbers, and neither is the site's.
What would move the score. More public model-level findings.
- Funding 3 to 4: score –
- Governance 2 to 3: score –
- Personnel 2 to 3: score –
Dissent, lower. Publication should be 1. RAND's frontier-model work reaches the public mainly through METR's Canary summaries and government-restricted reports; no RAND-authored adverse finding about a named lab's model is on record, and the RAND-Irregular model-theft paper is generic. Findings that surface only as another organization's summaries are R.3's lab-summarized citations in all but name.
Dissent, higher. Publication should be 3. RAND's research-integrity page commits to free and open publication of findings, disclosure of every funding source, and policies for intellectual independence; thousands of reports are free on rand.org. What is withheld is classified by government, not redacted by any lab, and R.4 tags that mechanism statutory rather than as lab control.
Each value is the tightest admissible cap or, with no cap, the highest admissible floor, under RULES.md. The binding signal is highlighted. A dash means no admissible signal sets a bound under the Retrieved & confirmed policy.
- Large share of AI work funded by Coefficient Giving. caps at 3 by F.6binding“Board of directors Dustin Moskovitz, Cari Tuna, Divesh Makan, Holden Karnofsky, and Alexander Berger”
RAND and the AI Evaluator Forum (RAND, 2026) tier 3 self (rand)confirmed Coefficient Giving (Wikipedia, 2026) tier 4 press (aggregator)confirmed Anthropic raises $124 million Series A (Anthropic, 2021-05-28) tier 3 self (anthropic)confirmed evaluators.csv (24 evaluators: Coefficient and SFF totals, lab money, government contracts, leadership) (Kevin Bass (GitHub), 2026-09-14) tier 2 ledgerimported2026-09-14 - FY2024 revenue of about $489M is mostly US government contract research; Coefficient's $87M since 2020 is a small share. floors at 3 by F.9“Contributions $488,777,692 91.4%”
evaluators.csv (24 evaluators: Coefficient and SFF totals, lab money, government contracts, leadership) (Kevin Bass (GitHub), 2026-09-14) tier 2 ledgerimported Rand Corporation, Form 990 FY ending Sept. 2025 (Nonprofit Explorer) (ProPublica, 2026) tier 1 filingconfirmed2026-09-11 - The Audacious commitment to Canary is about $38 million across RAND and METR; RAND's Coefficient grants include $10.5M for emerging technology initiatives. floors at 3 by F.6“Canary—a Collaboration with METR—to Receive Approximately $38 Million”
RAND AI Security Project Receives Funding Commitment Through The Audacious Project (RAND, 2024-10-09) tier 3 self (rand)confirmed Coefficient Giving grants index (Coefficient Giving, 2026-09-11) tier 1 indeximported2024-10-09 - SFF recommended $1,022,000 to RAND's Technology and Security Policy Center in 2025; SFF's funder Jaan Tallinn led Anthropic's Series A. caps at 3 by F.6binding“RAND Corporation [Technology and Security Policy Center] Main: $274,000 Freedom: $749,000”
SFF-2025 S-Process Recommendations Announcement (Survival and Flourishing Fund, 2025) tier 3 self (sff)confirmed Anthropic raises $124 million Series A (Anthropic, 2021-05-28) tier 3 self (anthropic)confirmed2026-09-15
- Government-contracted with long-standing COI and publication norms; AEF founding member. floors at 2 by G.1binding“a policy of mandatory disclosure”
RAND and the AI Evaluator Forum (RAND, 2026) tier 3 self (rand)confirmed AI Evaluator Forum launch and AEF-1 (AI Evaluator Forum, 2025-12-04) tier 3 self (aef)confirmed Research Integrity (RAND, 2026) tier 3 self (rand)unaudited2026-09-14
- Institutional COI rules. floors at 2 by P.6binding“a policy of mandatory disclosure”
RAND and the AI Evaluator Forum (RAND, 2026) tier 3 self (rand)confirmed Research Integrity (RAND, 2026) tier 3 self (rand)unaudited2026-09-14
- Access through government channels; defense research passes an Air Force public-release clearance gate. floors at 3 by A.4binding“This report was cleared for public release by the Department of the Air Force on September 13, 2023.”
RAND and the AI Evaluator Forum (RAND, 2026) tier 3 self (rand)confirmed The Cost of the Ukraine War for Russia (RAND RRA2421-1) (RAND, 2023) tier 3 self (rand)confirmed2026-09-14
- Sets own research agenda. floors at 3 by S.3binding“This project is a RAND Venture. Funding was provided by gifts from RAND supporters and income from operations.”
RAND and the AI Evaluator Forum (RAND, 2026) tier 3 self (rand)confirmed Algorithmic Equity: A Framework for Social Applications (RAND, 2019) tier 3 self (rand)confirmed2026-09-14
- Much output is government-restricted rather than public. caps at 2 by R.4
RAND and the AI Evaluator Forum (RAND, 2026) tier 3 self (rand)confirmed2026-09-14 - Thousands of RAND studies are available to the public for free on rand.org. floors at 3 by R.4not counted under standard“Thousands of RAND studies are available to the public for free on rand.org.”
Research Integrity (RAND, 2026) tier 3 self (rand)unaudited2026-09-15
- Methods partly public. caps at 2 by M.2binding“We explain our research, analysis, findings, and recommendations in ways that are understandable and usable.”
RAND and the AI Evaluator Forum (RAND, 2026) tier 3 self (rand)confirmed RAND Standards for High-Quality and Objective Research and Analysis (RAND, 2022) tier 3 self (rand)confirmed2026-09-14
- No commercial products. floors at 4 by X.7not counted under standard“RAND is nonprofit and nonpartisan.”
Research Integrity (RAND, 2026) tier 3 self (rand)unaudited2026-09-14
Values under each evidence policy
| Dimension | Leads included | Retrieved & confirmed | Against interest | Verified spans | Primary only |
|---|---|---|---|---|---|
| Funding | 3 | 3 | 3 | 3 | 3 |
| Governance | 2 | 2 | 2 | 2 | – |
| Personnel | 2 | 2 | – | 2 | – |
| Access depth (lab-granted) | 3 | 3 | – | 3 | – |
| Scope control | 3 | 3 | – | 3 | – |
| Publication rights | 3 | – | – | – | – |
| Method transparency | 2 | 2 | 2 | 2 | – |
| Role incompatibility | 3 | – | – | – | – |
| Score, Lab procurement | – No floor triggered, 8/8 | – No floor triggered, 6/8 | – No floor triggered, 3/8 | – No floor triggered, 6/8 | – No floor triggered, 1/8 |
| Score, Regulator or auditor | – No floor triggered, 8/8 | – No floor triggered, 6/8 | – No floor triggered, 3/8 | – No floor triggered, 6/8 | – No floor triggered, 1/8 |
| Score, Public trust | – No floor triggered, 8/8 | – No floor triggered, 6/8 | – No floor triggered, 3/8 | – No floor triggered, 6/8 | – No floor triggered, 1/8 |
| Score, Equal weights | – No floor triggered, 8/8 | – No floor triggered, 6/8 | – No floor triggered, 3/8 | – No floor triggered, 6/8 | – No floor triggered, 1/8 |
Traced money and ties
hop 2: 1 row (recommendation $1.0M); unattributed: 3 rows (commitment $38.0M). Confirmed rows: 4 of 4 (3 excluded after a re-fetch disagreed or could not be verified: T02, T58, T59) Dollar sums: confirmed + unaudited USD rows only; imported figures are not re-derived and never summed. Second hop traced for 2 of 4 sources: untraced High Tide Foundation, Valhalla Foundation. Ties: none within two steps recorded. Checked and not found: 0.
Ledger
Money in
- T02 from Coefficient Giving (formerly Open Philanthropy): grant $10.00M; quarantined — excluded from sums, 2025-09-20. AI Evaluation and Testing unverifiabletier 1 indexhttps://coefficientgiving.org/grants/ai-evaluation-and-testing/
- T27 from The Audacious Project (TED): commitment $38.00M, 2024-10-09. Canary commitment to RAND with METR as partner confirmedtier 3 selfhttps://www.rand.org/news/press/2024/10/09.html
- T58 from Coefficient Giving (formerly Open Philanthropy): grant $76.96M; quarantined — excluded from sums, 2020-2025. cumulative Coefficient grants to RAND excluding the $10M in T02 unverifiabletier 1 indexhttps://coefficientgiving.org/
- T59 from US government (NIST appropriations): contract $488.78M; quarantined — excluded from sums, FY2024. FY2024 revenue, mostly federal contract research unverifiabletier 1 filinghttps://projects.propublica.org/nonprofits/organizations/951958142
- T70 from Survival and Flourishing Fund: recommendation $1.02M, 2025. SFF 2025: $1,022,000 to the Technology and Security Policy Center confirmedtier 3 selfhttps://survivalandflourishing.fund/2025/recommendations
- T102 from Valhalla Foundation: grant $10.00M; detail of T27 — not additive, 2024. Project Canary, an artificial intelligence safety initiative (990-PF TY2024 Part XV) confirmedtier 1 filinghttps://gt990datalake-rawdata.s3.amazonaws.com/EfileData/XmlFiles/202502559349100000_public.xml
- T103 from High Tide Foundation: grant $0.33M; detail of T27 — not additive, 2024. 'to support the Project Canary' as filed (990-PF TY2024 Part XV) confirmedtier 1 filinghttps://gt990datalake-rawdata.s3.amazonaws.com/EfileData/XmlFiles/202503179349100135_public.xml
Funding graph, focused
Every one-way chain to a lab and to an evaluator at full strength, everything else faded. Hover any name to trace its connections; click a name to open its page.