SaferAI
Scorecard
Rates lab risk-management frameworks in public; runs EU Code of Practice evaluations.
Nonprofit, Paris, FR. Independent referee; listed with independent referees. Confidence med. Domains: assurance, cyber, bio, 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. Documented pre-release access under Code terms.
- Funding 3 to 4: score –
- Governance 2 to 3: score –
- Personnel 2 to 3: score –
Dissent, lower. Access could be 0. No source records any pre-release model access for SaferAI: the goal-directedness evaluation it co-authored with Google DeepMind researchers tested models from Google DeepMind, OpenAI and Anthropic through ordinary means, and its own site describes framework ratings, risk models and advisory work rather than model testing. Anchor 0 is public API only, and that is the whole documented record.
Dissent, higher. Access could be 2. SaferAI leads the risk-modelling workstream of the EU AI Office's CBRN lot under a three-year contract, participated in all four Code of Practice working groups, and co-authors with a frontier developer's researchers; the Code contemplates external evaluator access to models before release. Documented pre-release access under Code terms, which the card names as what would move the score, would reach anchor 1 or 2.
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.
- No lab revenue found. floors at 2 by F.10
SaferAI (SaferAI, 2026) tier 3 self (saferai)confirmed2026-09-14 - SFF recommendations of about $1.06M over 2022 to 2025; EU AI Office Lot 1 consortium member. floors at 3 by F.12“SaferAI Main: $0 Freedom: $307,000 Fairness: $4,000 Mean: $0 $311,000”
evaluators.csv (24 evaluators: Coefficient and SFF totals, lab money, government contracts, leadership) (Kevin Bass (GitHub), 2026-09-14) tier 2 ledgerimported TED 864574-2025 Technical Assistance for AI Safety (Tenders Electronic Daily, 2025-12) tier 1 filingconfirmed SFF-2025 S-Process Recommendations Announcement (Survival and Flourishing Fund, 2025) tier 3 self (sff)confirmed2026-09-11 - SFF recommended $311,000 to SaferAI in 2025 (with a $111,000 match) after earlier rounds; SFF's funder is Jaan Tallinn, the Anthropic Series A lead. caps at 3 by F.6binding“SaferAI Main: $0 Freedom: $307,000 Fairness: $4,000 Mean: $0 $311,000”
SFF-2025 S-Process Recommendations Announcement (Survival and Flourishing Fund, 2025) tier 3 self (sff)confirmed2026-09-15
- Nonprofit. floors at 2 by G.1binding“SaferAI, a registered non-profit (reg no. 919 185 199 00016”
SaferAI (SaferAI, 2026) tier 3 self (saferai)confirmed2026-09-14
- No lab roles found. floors at 2 by P.6binding“We're hiring people who want to shape how AI is developed.”
SaferAI (SaferAI, 2026) tier 3 self (saferai)confirmed2026-09-14
- Limited pre-deployment model access to date. caps at 1 by A.1binding
SaferAI (SaferAI, 2026) tier 3 self (saferai)confirmed2026-09-14
- Part of the EU AI Office evaluation consortium; sets its own rating methodology. floors at 3 by S.5“We actively shaped the EU's Code of Practice for general-purpose AI models”
FAR AI (Longterm Wiki, 2026-02-26) tier 4 press (aggregator)confirmed SaferAI (SaferAI, 2026) tier 3 self (saferai)confirmed2026-09-14 - SaferAI's executive director and founder co-authored "Evaluating the Goal-Directedness of Large Language Models" (April 2025) with Google DeepMind researchers, an evaluation of models from Google DeepMind, OpenAI and Anthropic, while SaferAI publicly rates Google DeepMind's risk-management framework. caps at 3 by S.6binding“Authors: Tom Everitt , Cristina Garbacea , Alexis Bellot , Jonathan Richens , Henry Papadatos , Siméon Campos”
Evaluating the Goal-Directedness of Large Language Models (arXiv:2504.11844) (arXiv, 2025-04-16) tier 4 press (aggregator)confirmed About (SaferAI, 2026) tier 3 self (saferai)confirmed2026-09-15
- Publishes critical ratings of lab frameworks with scores attached. floors at 2 by R.7binding“We independently evaluate and rate how well leading AI companies manage risks”
SaferAI (SaferAI, 2026) tier 3 self (saferai)confirmed Top AI Companies Have ‘Unacceptable’ Risk Management, Studies Say (TIME, 2025-07-17) tier 4 press (primary)confirmed2026-09-14
- Rating methodology published. floors at 2 by M.2binding“how they identify risks, assess their severity, implement safeguards, and govern their systems”
SaferAI (SaferAI, 2026) tier 3 self (saferai)confirmed2026-09-14
- No commercial products. floors at 3 by X.4
SaferAI (SaferAI, 2026) tier 3 self (saferai)confirmed2026-09-14 - Advisory work for G42 and for an unnamed major AI company, alongside public framework ratings; no lab-money policy on the site. caps at 2 by X.3not counted under standard“advised G42 and 'a major AI company' on frontier safety frameworks”
evaluators.csv (24 evaluators: Coefficient and SFF totals, lab money, government contracts, leadership) (Kevin Bass (GitHub), 2026-09-14) tier 2 ledgerimported2026-09-11 - SaferAI says it contributed to frameworks for G42 and a major AI company and co-drafted G42's Frontier AI Safety policy; G42 names METR and SaferAI as supporting experts. caps at 2 by X.3binding“We've contributed to frameworks for G42 and a major AI company”
About (SaferAI, 2026) tier 3 self (saferai)confirmed2026-09-15 - Document request: Identity of the major AI company client.
- Document request: A published lab-money policy.
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 | – | – |
| Scope control | 3 | 3 | 3 | 3 | – |
| Publication rights | 2 | 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 0: 1 row; hop 2: 1 row (recommendation $0.3M); public: 1 row. Confirmed rows: 2 of 3 (1 excluded after a re-fetch disagreed or could not be verified: T40) Dollar sums: confirmed + unaudited USD rows only; imported figures are not re-derived and never summed. Second hop traced for 3 of 3 sources. Ties: none within two steps recorded. Checked and not found: 1.
Ledger
Money in
- T40 from Survival and Flourishing Fund: recommendation $0.75M; quarantined — excluded from sums, 2022-2024. SFF recommendations 2022 to 2024 (cumulative minus the confirmed 2025 row) differstier 3 selfhttps://survivalandflourishing.fund/
- T41 from EU budget (Digital Europe Programme): contract €1.43M (EUR, not converted), 2025-12-04. EU AI Office Lot 1 (CBRN): consortium lot total EUR 1,434,080 led by FAR.AI with SecureBio and SaferAI; SaferAI's share not published confirmedtier 3 selfhttps://ted.europa.eu/en/notice/-/detail/864574-2025
- T68 from Survival and Flourishing Fund: recommendation $0.31M, 2025. SFF 2025: $311,000 incl. $90,000 speculation and $111,000 matching confirmedtier 3 selfhttps://survivalandflourishing.fund/2025/recommendations
- T84 from G42: contract undisclosed, 2025. advisory on G42's Frontier AI Safety Framework; terms undisclosed unauditedtier 3 selfhttps://www.g42.ai/resources/news/g42-publishes-its-frontier-ai-safety-framework
Checked and not found
- N07 Does Coefficient Giving fund SaferAI?
Searched Coefficient grants index, 2,911 rows, snapshot 2026-09-11: no grant from Coefficient Giving in its index. Prompted by imported lead: kevinnbass ledger row EV10. importedhttps://coefficientgiving.org/grants/
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.