Evaluators

26 ranked organizations scored on eight independence dimensions, split by role (referee, government, vendor, benchmark, lab team). Columns show the weighted score under each preset, the weakest dimension, confidence, and how many ledger inflow rows are confirmed. Open a row for the full scorecard, ledger, and focused graph.

Evaluator	Role	Lab	Regulator	Public	Equal	Floor	Confidence	Confirmed rows	Domains
METR Nonprofit, Berkeley, US	referee	81	80	75	81	personnel 2	high	5/7	autonomy, scheming, incident, assurance
AVERI Nonprofit, Washington, US	referee	79	78	78	81	funding 3	med	4/4	assurance
SaferAI Nonprofit, Paris, FR	referee	76	79	80	75	access depth (lab-granted) 2	med	2/3	assurance, cyber, bio, misuse
EU AI Office Government, Brussels, BE	government	75	74	73	75	publication rights 2	med	1/1	bio, cyber, misuse, assurance
FAR.AI Nonprofit, Berkeley, US	referee	75	75	75	75	funding 3	high	4/6	jailbreak, bio, cyber, misuse
UK AI Security Institute Government, London, UK	government	75	74	73	78	publication rights 2	high	2/3	cyber, bio, jailbreak, autonomy, assurance
Redwood Research Nonprofit, Berkeley, US	referee	74	73	73	75	personnel 2	med	1/2	scheming, incident, assurance
Transluce Nonprofit, Berkeley, US	referee	73	75	73	75	funding 2	high	3/3	misuse, autonomy, benchmarks, assurance
RAND Nonprofit, Santa Monica, US	referee	71	69	70	72	publication rights 2	med	2/2	bio, cyber, assurance
SecureBio Nonprofit, Cambridge, US	referee	71	71	69	72	funding 2	high	3/3	bio, misuse
Apollo Research Public benefit corp, London, UK / San Francisco, US	referee	67	69	64	66	funding 2	high	2/2	scheming, autonomy, assurance
Epoch AI Nonprofit, Remote	referee	64	65	64	69	funding 2	high	4/8	benchmarks
Andon Labs VC-backed, Stockholm, SE / San Francisco, US	referee	60	61	59	59	funding 2	low	1/1	autonomy, benchmarks
Center for AI Safety Nonprofit, San Francisco, US	referee	60	60	60	60	personnel 1	med	2/2	benchmarks, misuse
US CAISI (NIST) Government, Gaithersburg, US	government	60	60	60	60	publication rights 1	high	1/2	cyber, bio, misuse, assurance
Holistic Agent Leaderboard (Princeton) Academic, Princeton, US	benchmark	60	60	60	60	access depth (lab-granted) 1	med	0/0	benchmarks, autonomy
Humane Intelligence Nonprofit, New York, US	referee	60	60	60	60	access depth (lab-granted) 1	med	0/0	misuse, jailbreak
Palisade Research Nonprofit, Berkeley, US	referee	60	60	60	60	access depth (lab-granted) 1	med	1/2	cyber, autonomy, scheming
EquiStamp Private company, capital structure undisclosed, Remote	referee	59	59	59	56	governance 2	low	1/1	assurance, autonomy
MLCommons (AILuminate) Consortium, San Francisco, US	benchmark	59	60	55	60	funding 1	med	1/1	benchmarks, assurance
Irregular (formerly Pattern Labs) VC-backed, Tel Aviv, IL / San Francisco, US	vendor	47	49	40	44	funding 1	high	2/2	cyber, misuse
Nemesys Insights Private company, capital structure undisclosed, Albany, US	vendor	46	46	44	47	funding 1	low	2/2	bio, misuse, assurance
Dreadnode VC-backed, Remote	vendor	42	43	34	38	funding 1	low	0/0	cyber
Microsoft AI Red Team Big Tech unit, Redmond, US	lab-team	41	43	31	38	funding 1	med	1/1	jailbreak, cyber, misuse
Gray Swan VC-backed, Pittsburgh, US	vendor	40	40	35	38	role incompatibility 0	high	2/2	jailbreak, cyber, misuse
Scale AI (SEAL / Scale Labs) VC-backed, San Francisco, US	vendor	38	40	26	34	funding 0	high	1/1	benchmarks, cyber, bio, misuse

Watchlist (not ranked)

Expected entrants scored on the same rubric but excluded from rankings and averages: hypothetical composites and announced initiatives with no evaluations yet.

Evaluator	Role	Lab	Regulator	Public	Equal	Floor	Confidence	Confirmed rows	Domains
Assurance firms (Big Four and peers) Hypothetical composite, not a real organization, Global	expected-entrant	38	39	41	38	access depth (lab-granted) 1	low	0/0	assurance
Hugging Face Open Alignment Initiative VC-backed, New York, US / Paris, FR	expected-entrant	37	38	40	40	governance 0	low	1/1	assurance