SaferAI
- Type
- Watchdog or incident database
- Lieu
- France — France (Paris, 6 Rue d'Armaillé, 75017)
- Dernière vérification
- 2026-09-22
- Prochaine vérification
- 2027-03-21
Pas encore traduit — affiché en anglais.
Comment les joindre
- General enquiries
- Company risk-management ratings portal
- Homepage
- Action page
Ce canal est conservé uniquement comme référence. Sa vérification est absente, date de plus de 180 jours, ou la période de contact n'est pas ouverte actuellement ; ne vous y fiez pas avant une nouvelle vérification.
Ce qu'il fait
Non-profit focused on AI risk management. Three arms: research (quantitative risk models — nine cyber risk models completed, expanding to loss of control and CBRN); ratings (independent grading of how well leading AI companies manage risk, across risk identification, severity assessment, safeguards and governance, published at ratings.safer-ai.org); and standards (leads AI risk management standards work in the EU and edits international red-teaming standards).
Ce qu'ils demandent — dans leurs propres termes
AI companies should be held to real risk-management standards, and governance should be built on rigorous, auditable frameworks rather than voluntary promises.
Ce qu'ils font
Research on risk management frameworks, public ratings of AI companies' safety practices (ratings.safer-ai.org), and standards work in EU and international governance processes.
Évaluation franche
Small, technically serious and European — which matters, because it has standing in the EU standardisation processes that most US safety organisations do not. Responsive to substantive technical input on risk methodology.
Ce qu'un nouveau venu peut faire
Read and cite the company ratings when arguing with policymakers or employers; donate; apply for an open role.
Bon à savoir
A research and standards body, not a campaign — there is no membership or activist route. Its ratings are useful ammunition for other people's campaigns.
Préoccupations couvertes
Risk-management adequacy at frontier developers — explicitly covering cyber, loss of control and CBRN. Methodologically the most rigorous of the company-rating projects, borrowing from mature safety-critical industries.
Canal gouvernemental
Real in Europe — SaferAI staff serve as editors on international standards and participate in EU AI Act standardisation and the GPAI Code of Practice process.
Conseils
unclear
Où ils agissent
France (Paris) and EU-level; registered French nonprofit, reg. no. 919 185 199 00016.
Taille
Small technical nonprofit; headcount not published.
Dépenses
Not disclosed.