Expertise Rooted in French TrustworthyAI Deeptech
AICET is designed and operated by Numalis, a French deeptech company recognized for its expertise in the reliability of high-risk artificial intelligence. Scientific rigor transferred to AI skills measurement, to provide large organizations with an enforceable benchmark.
Numalis: a French deeptech company
specializing in trustworthy AI
Born out of scientific research, Numalis specialises in the explainability and robustness validation of AI using formal methods. This branch of theoretical computer science leverages mathematical logic to prove the validity of the information produced by an AI.
Unlike traditional testing, which is limited to statistical evaluations, Numalis explores the entire realm of possibilities of an AI model using its Saimple software. The objective: to detect the maximum level of disturbance at which the model's reliability is guaranteed.
From Saimple to AICET: one conviction, two objects
By supporting the largest industrial players with the technical validation of their algorithms for MORE than a decade, Numalis has identified a major blind spot: the reliability of the human skills operating these systems. Securing the code is no longer enough; adoption must also be secured.
AICET was born from this observation: applying to AI skills measurement the same scientific and mathematical rigor that Saimple applies to model validation. One conviction: you cannot steer an AI transformation without a rigorous, auditable measurement neither of the systems OR EVEN of the skills.
Numalis Milestones
Saimple
Proprietary software for formal analysis of AI models
€5M funding round
(November 2023)
Definvest (Ministry of Defence / BPI France), La Banque Postale, MBDA, Safran Corporate Ventures
2026
Renewed support from 574 Invest, the SNCF Group's investment fund
Key partners
SNCF (7 years), DGA, MBDA (since 2019), Safran, La Banque Postale, BPI France
From scoring to AI Strategy
Holding a PhD in AI, Arnault Ioualalen is a leading figure in trustworthy AI in France. Founder of Numalis in 2015 and subsequently of AICET, he holds the conviction that the reliability of AI systems and of the human skills that operate them stems from the same scientific requirement: measurable, enforceable, aligned with standards.
His normative commitment is concrete and public: he is the lead editor of the international standard ISO/IEC 24029 on AI robustness, a direct contributor to AFNOR SPEC 2401 (the French reference framework for measuring AI skills) and to the AI Act, an international expert on ISO/IEC trustworthy AI committees, and an active AFNOR member.
« Notre mission chez Numalis est simple : transformer l'incertitude liée à l'IA en une fiabilité absolue, vérifiable et conforme aux exigences de demain. »
Standards Commitments
ISO/IEC committees
on trustworthy AI
International expert
AFNOR
Active member of standardization committees
Numalis Achievements — Three flagship cases
Numalis's experience in validating critical AI has been built alongside major players in regulated sectors. Three illustrations.
Challenge
Validate the reliability of AI vision algorithms for the dashboard displays of the European ERTMS signaling system, in a domain where safety is an absolute requirement. The deep learning ‘black box’ effect renders traditional certification methods inoperable.
Approach
Application of Saimple software (formal methods) to mathematically explore the model’s space of possibilities; injection of perturbations (blur, pixel modifications, watermarks) to detect invisible vulnerabilities.
Result
Automation of ERTMS DMI screen certification, major time savings in the validation phase, 7-year operational partnership, equity participation from 574 Invest (SNCF Group fund) in 2026. Numalis becomes a building block of the global AI railway certification standard.
Challenge
Gradually integrate AI algorithms into operational equipment while guaranteeing integrity, safety, and explainability — non-negotiable requirements in a national sovereignty context.
Approach
Development of supervision and algorithmic control tools based on formal methods. Initial defence-classified programs on missile reliability, then extension to embedded AI systems (MBDA partnership since 2020).
Result
Numalis recognized as a strategic technology asset by the Ministry of Defence, multi-year industrial partnership with MBDA, equity support from the Definvest fund (BPI France) and MBDA directly.
Challenge
Facilitate AI adoption in a sector subject to some of the world's most rigorous safety standards, where every embedded algorithm must be certifiable and explainable.
Approach
Collaboration on explainability and algorithmic validation solutions adapted to critical aeronautical environments, leveraging Numalis’s proprietary formal methods.
Result
Equity participation from Safran Corporate Ventures in the €5M fundraising round of November 2023. Numalis identified as a strategic partner for the industrialization of trustworthy AI in aeronautics.
Our Methodological Rigor
Advanced psychometricsMeticulous item calibration to guarantee reliable, unbiased, and repeatable results.
Standards alignmentA method in direct alignment with the AFNOR (SPEC 2401), ISO/IEC (including standard 24029), and IEC frameworks.
Continuous validationRegular scientific reviews combined with field feedback from large partner organizations.
Recognition & commitments
€5M fundraising round (2023)
Definvest (BPI France), 115K (La Banque Postale), MBDA, Safran Corporate Ventures
574 Invest
2026 participation from the SNCF Group's investment fund
Active AFNOR member and contributor to SPEC 2401
ISO/IEC contributor
Lead editor of the ISO/IEC 24029 standard (AI robustness)
France 2030
/ France Relance programs
+75% of PhD researchers in R&D
Academic partnerships
University of Montpellier
Our conviction
Measuring AI skills is an act of scientific engineering, not a communication exercise. Designed and operated by a deeptech company recognized for the reliability of critical AI: AICET is a standard, not just a barometer.
Entrust Your AI Skills Measurement
to a Team of Scientists
