evaluator; distance from a lab: 1
FAR.AI
Scorecard
Red-teaming and adversarial-robustness nonprofit; leads the EU AI Office CBRN risk-modelling contract.
Nonprofit, Berkeley, US. Independent referee; listed with independent referees. Confidence high. Domains: jailbreak, bio, cyber, misuse.
No floor triggered 8/8 evidenced · Retrieved & confirmed Scores by preset: Lab procurement –; Regulator or auditor –; Public trust –; Equal weights –. Weakest evidenced dimension: access depth (lab-granted) 1/4. No evidenced dimension below 2 — not a pass. No floor triggered means no disqualifying conflict is on file, not a pass. Full evidence.
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. Disclose the AI Safety Fund share of budget and whether lab-funded pools are excluded from evaluation work.
- Funding 3 to 4: score –
- Governance 2 to 3: score –
- Personnel 2 to 3: score –
Dissent, lower. Access could be 0. No cited source records the terms of any lab engagement: the wiki says FAR.AI began red-teaming leading models for frontier labs in late 2023 and that publicly confirmed engagements include OpenAI, but neither it nor the $30M post shows pre-release checkpoints, safeguards off, or logs. Anchor 0 is public API only, and that is all the record demonstrates; the EU contract is a regulator engagement, not lab access.
Dissent, higher. Access could be 2. OpenAI's GPT-5 system card credits FAR.AI with 80 hours of red-teaming before release, and pre-release red-teamers typically test early checkpoints under relaxed safeguards; the transparency page describes engagements with companies building frontier models under NDA. If the engagement terms show safeguards off or extended windows, anchor 2 applies, and helpful-only or log access would reach 3.
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.
Funding 3/4
Anchor 3: Mostly philanthropic or public money; some lab-linked pooled funds.
evidence: tier 1 (filing/index)
Capped at 3 by farai.02 (F.6): Also funded by the AI Safety Fund, financed by the Frontier Model Forum, the labs' industry body.; capped at 3 by farai.12 (F.6): Largest single grant: $28,675,000 over three years from Coefficient Giving (September 2025); over $30 million in 2025 commitments from Coefficient, Schmidt Sciences, SFF, CSET and the Frontier Model Forum's AI Safety Fund.; capped at 3 by farai.17 (F.6): Founders Pledge, a donor-advised giving vehicle, granted $500,000 to the FAR.AI Integrity Team for EU AI Office manipulation evaluations.; capped at 3 by farai.16 (F.3): FAR.AI charges for-profit AI developers market rates for consulting and caps that revenue at 10% of total annual revenue; the transparency page reports TY2024 earned revenue of $429K against $24.3M total.. Floors up to 3 from farai.01, farai.10 do not exceed the cap. Not counted under this policy: farai.14 (no confirmed source (unaudited)).
- More than $30M in 2025 commitments from Coefficient Giving, Schmidt Sciences, SFF and CSET. floors at 3 by F.12“FAR.AI has secured over $30 million in funding commitments throughout 2025”
FAR.AI Secures Over $30 Million in Multi-Funder Support (FAR.AI, 2026-07-16) tier 3 self (farai)confirmed2026-09-14 - Also funded by the AI Safety Fund, financed by the Frontier Model Forum, the labs' industry body. caps at 3 by F.6binding“the AI Safety Fund (AISF), supported by the Frontier Model Forum (FMF)”
FAR.AI Secures Over $30 Million in Multi-Funder Support (FAR.AI, 2026-07-16) tier 3 self (farai)confirmed2026-09-14 - Published cap: revenue from for-profit AI developers limited to 10% of annual revenue, charged at market rates with publication rights retained; TY2024 earned revenue $429K of $24.3M. floors at 3 by F.3“is capped at a maximum of 10% of FAR.AI's total annual revenue”
Transparency (FAR.AI, 2026) tier 3 self (farai)confirmed2026-09-14 - Largest single grant: $28,675,000 over three years from Coefficient Giving (September 2025); over $30 million in 2025 commitments from Coefficient, Schmidt Sciences, SFF, CSET and the Frontier Model Forum's AI Safety Fund. caps at 3 by F.6binding“FAR.AI has secured over $30 million in funding commitments throughout 2025”
FAR.AI General Support (2022) (Open Philanthropy / Coefficient Giving, 2022) tier 1 indexunaudited FAR.AI Secures Over $30 Million in Multi-Funder Support (FAR.AI, 2026-07-16) tier 3 self (farai)confirmed2025-09 - A third-party wiki records a 10% cap on for-profit developer revenue. FY2024 revenue of $24.3M against $8.6M expenses is per the same source; span not yet captured. floors at 3 by F.3not counted under standard“caps revenue from for-profit AI developers at 10% of total annual revenue”
FAR AI (Longterm Wiki, 2026) tier 2 ledgerunaudited2026-09-15 - Founders Pledge, a donor-advised giving vehicle, granted $500,000 to the FAR.AI Integrity Team for EU AI Office manipulation evaluations. caps at 3 by F.6binding“FAR.AI Integrity Team ($500,000)”
Frontier AI grantmaking (Founders Pledge, 2026) tier 3 self (founders-pledge-frontier)confirmed2026-09-15 - FAR.AI charges for-profit AI developers market rates for consulting and caps that revenue at 10% of total annual revenue; the transparency page reports TY2024 earned revenue of $429K against $24.3M total. caps at 3 by F.3binding“is capped at a maximum of 10% of FAR.AI's total annual revenue”
Transparency (FAR.AI, 2026) tier 3 self (farai)confirmed2026-09-15
Governance 2/4
Anchor 2: For-profit or PBC with a published COI policy.
evidence: tier 3 (self)
Floored at 2 by farai.05 (G.1): Nonprofit with diversified disclosed funders.. No admissible signal caps it.
- Nonprofit with diversified disclosed funders. floors at 2 by G.1binding“FAR.AI is an independent nonprofit research organization”
FAR.AI Secures Over $30 Million in Multi-Funder Support (FAR.AI, 2026-07-16) tier 3 self (farai)confirmed2026-09-14
Personnel 2/4
Anchor 2: Frequent two-way hiring; recusal on request.
evidence: tier 4 (press)
Capped at 2 by farai.07 (P.3): Runs field-building programs that place researchers into labs..
- Runs field-building programs that place researchers into labs. caps at 2 by P.3binding
FAR AI (Longterm Wiki, 2026-02-26) tier 4 press (aggregator)confirmed2026-09-14
Access depth (lab-granted) 1/4
Anchor 1: Pre-release API with safeguards on.
evidence: tier 4 (press)
mechanism: lab-controlled
Floored at 1 by farai.06 (A.1): Pre- and post-deployment red teaming for labs and EU bodies.. No admissible signal caps it.
- Pre- and post-deployment red teaming for labs and EU bodies. floors at 1 by A.1binding“FAR AI began red-teaming leading language models for frontier labs in Q4 2023, including red-teaming of GPT-4”
FAR AI (Longterm Wiki, 2026-02-26) tier 4 press (aggregator)confirmed2026-09-14
Scope control 3/4
Anchor 3: Evaluator sets scope and can add questions.
evidence: tier 1 (filing/index)
Floored at 3 by farai.03 (S.3): Selected by the European Commission to lead a three-year CBRN technical-assistance contract, with a regulator as client.; floored at 3 by farai.11 (S.3): Leads EU AI Office Lot 1 (CBRN) with SecureBio and SaferAI, EUR 1.43M to the consortium.. No admissible signal caps it.
- Selected by the European Commission to lead a three-year CBRN technical-assistance contract, with a regulator as client. floors at 3 by S.3binding“FAR AI was selected by the European Commission's AI Office to lead Lot 1 (CBRN Risk Modelling and Evaluation)”
FAR AI (Longterm Wiki, 2026-02-26) tier 4 press (aggregator)confirmed2026-09-14 - Leads EU AI Office Lot 1 (CBRN) with SecureBio and SaferAI, EUR 1.43M to the consortium. floors at 3 by S.3binding
TED 864574-2025 Technical Assistance for AI Safety (Tenders Electronic Daily, 2025-12) tier 1 filingconfirmed2026-09-14 - Founders Pledge granted $500,000 to the FAR.AI Integrity Team for EU AI Office manipulation evaluations. informational (Section 9 (mis-dimensioned facts): a grant is a funding fact; informational on S and re-recorded on F as farai.17 (F.6). The regulator-client fact is already carried by farai.03 and farai.11.)“The grant funds the evaluations FAR.AI is now developing for the EU AI Office's consortium on manipulation”
Frontier AI grantmaking (Founders Pledge, 2026) tier 3 self (founders-pledge-frontier)confirmed2026
Publication rights 2/4
Anchor 2: Publishes; lab reviews with broad redaction.
evidence: tier 3 (self)
mechanism: self-imposed
Floored at 2 by farai.04 (R.7): Publishes jailbreak and stress-test results across vendors.. No admissible signal caps it.
- Publishes jailbreak and stress-test results across vendors. floors at 2 by R.7binding“we've published groundbreaking research on adversarial robustness, interpretability, and red-teaming”
FAR.AI Secures Over $30 Million in Multi-Funder Support (FAR.AI, 2026-07-16) tier 3 self (farai)confirmed2026-09-14
Method transparency 2/4
Anchor 2: Methods described in prose.
evidence: tier 3 (self)
Floored at 2 by farai.08 (M.2): Publishes methods and papers.. No admissible signal caps it.
- Publishes methods and papers. floors at 2 by M.2binding“we've published groundbreaking research on adversarial robustness, interpretability, and red-teaming”
FAR.AI Secures Over $30 Million in Multi-Funder Support (FAR.AI, 2026-07-16) tier 3 self (farai)confirmed2026-09-14
Role incompatibility 2/4
Anchor 2: Consults for labs.
evidence: tier 3 (self)
floor and cap disagree
Conflict: floor 3 from farai.09 (X.4) against cap 2 from farai.15 (X.3); resolved at 2 by X.3: FAR.AI's transparency page records paid work for for-profit frontier developers at market rates; paid consulting for a lab caps role incompatibility at 2 whatever else is open or free.
Resolution: X.3 decides 2. FAR.AI's transparency page records paid work for for-profit frontier developers at market rates; paid consulting for a lab caps role incompatibility at 2 whatever else is open or free.
- No commercial products; consulting limited to governments. floors at 3 by X.4“FAR AI conducts empirical technical AI safety research without developing or deploying AI products of its own”
FAR AI (Longterm Wiki, 2026-02-26) tier 4 press (aggregator)confirmed2026-09-14 - FAR.AI's transparency page publishes a consulting policy whose clients include companies building frontier AI models, charged at market rates, with revenue from for-profit developers capped at 10% of annual revenue; TY2024 earned revenue was $429K. caps at 2 by X.3binding“we charge market rates to for-profit developers to avoid subsidizing private actors”
Transparency (FAR.AI, 2026) tier 3 self (farai)confirmed2026-09-15
Values under each evidence policy
| Dimension | Leads included | Retrieved & confirmed | Against interest | Verified spans | Primary only |
|---|---|---|---|---|---|
| Funding | 3 | 3 | 3 | 3 | – |
| Governance | 2 | 2 | – | 2 | – |
| Personnel | 2 | 2 | 2 | – | – |
| Access depth (lab-granted) | 1 | 1 | 1 | 1 | – |
| Scope control | 3 | 3 | 3 | 3 | – |
| Publication rights | 2 | 2 | – | 2 | – |
| Method transparency | 2 | 2 | – | 2 | – |
| Role incompatibility | 2 | 2 | 2 | 2 | – |
| Score, Lab procurement | – No floor triggered, 8/8 | – No floor triggered, 8/8 | – No floor triggered, 5/8 | – No floor triggered, 7/8 | – Unevidenced, 0/8 |
| Score, Regulator or auditor | – No floor triggered, 8/8 | – No floor triggered, 8/8 | – No floor triggered, 5/8 | – No floor triggered, 7/8 | – Unevidenced, 0/8 |
| Score, Public trust | – No floor triggered, 8/8 | – No floor triggered, 8/8 | – No floor triggered, 5/8 | – No floor triggered, 7/8 | – Unevidenced, 0/8 |
| Score, Equal weights | – No floor triggered, 8/8 | – No floor triggered, 8/8 | – No floor triggered, 5/8 | – No floor triggered, 7/8 | – Unevidenced, 0/8 |
Traced money and ties
hop 1: 1 row; hop 2: 2 rows (recommendation $0.9M); public: 1 row; unattributed: 2 rows (grant $0.5M). Confirmed rows: 4 of 6 (2 excluded after a re-fetch disagreed or could not be verified: T03, T43) Dollar sums: confirmed + unaudited USD rows only; imported figures are not re-derived and never summed. Second hop traced for 5 of 6 sources: untraced Founders Pledge (frontier AI fund). Ties: none within two steps recorded. Checked and not found: 0.
Ledger
Money in
- T03 from Coefficient Giving (formerly Open Philanthropy): grant $59.35M; quarantined — excluded from sums, 2021-2025. cumulative grants, one funder one measure unverifiabletier 1 indexhttps://www.longtermwiki.com/wiki/E138
- T24 from AI Safety Fund (Frontier Model Forum): grant undisclosed, 2025. share of $30M+ multi-funder commitments, not itemised confirmedtier 3 selfhttps://far.ai/news/30m-multi-funder-support
- T42 from EU budget (Digital Europe Programme): contract €1.43M (EUR, not converted), 2025-12-04. EU AI Office Lot 1 CBRN lead; consortium total, EUR as written confirmedtier 3 selfhttps://www.far.ai/blog/far-ai-selected-to-lead-eu-ai-act-cbrn-risk-consortium
- T43 from Survival and Flourishing Fund: recommendation $5.59M; quarantined — excluded from sums, 2022-2024. SFF recommendations 2022 to 2024 (cumulative minus the confirmed 2025 row) unverifiabletier 3 selfhttps://survivalandflourishing.fund/recommendations
- T44 from Schmidt Sciences: grant undisclosed, 2025. named principal supporter confirmedtier 3 selfhttps://far.ai/news/30m-multi-funder-support
- T67 from Survival and Flourishing Fund: recommendation $0.92M, 2025. SFF 2025: $919,000 incl. $200,000 speculation and $719,000 matching at 0.25x confirmedtier 3 selfhttps://survivalandflourishing.fund/2025/recommendations
- T72 from Coefficient Giving (formerly Open Philanthropy): grant $28.68M; detail of T03 — not additive, 2025-09. largest single grant, three years unauditedtier 1 indexhttps://www.openphilanthropy.org/grants/far-ai-general-support/
- T76 from Founders Pledge (frontier AI fund): grant $0.50M, 2026. Integrity Team, EU AI Office manipulation evaluations unauditedtier 3 selfhttps://www.founderspledge.com/research/frontier-ai-grantmaking
Roles held
- R12 contractor at OpenAI, 2025. GPT-5 system card: FAR.AI conducted 80 hours of red-teaming; imported kevinnbass/metr-money-figure:B17 importedhttps://openai.com/index/gpt-5-system-card/
Roles hosted
- R11 Adam Gleave: founder, 2022. confirmedhttps://far.ai/
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.