Numalis and Octopize: advancing trustworthy AI with robust models and no re-identification risk

Nantes, 21 October 2025 – Numalis, a global leader in AI model robustness validation, and Octopize, a French deeptech specializing in synthetic anonymized data through its patented Avatar method, announce the results of a successful collaboration. The study demonstrates that AI models trained on synthetic anonymized data remain as performant and reliable as those trained on original datasets — while excluding the risk of re-identification or personal data leakage.

Concrete use cases

The objective of this joint work was to assess the impact of training AI models on avatar data — synthetic datasets generated from real data — in terms of robustness, compliance, and statistical fidelity.

The experiment focused on two use cases:

• A real-world scenario: heartbeat classification from electrocardiogram (ECG) time-series data, using the public Kaggle Heartbeat dataset.
• A use case with a client in Defense with strategic data: learning ML prediction models from avatar data.

With two main goals: guarantee privacy and security by excluding re-identification risks and validate robustness and statistical quality, ensuring that avatar-based models remain as effective as those trained on original data.

‍

Key findings

The results are clear and consistent:

• No re-identification risk → synthetic and anonymized data excludes the possibility of re-identifying individuals or leaking personal data.
• Performance preserved → Models trained on avatar data are as robust and accurate as those trained on original datasets.
• Statistical fidelity maintained → The same “uncertainty zones” were observed in both models, confirming equal predictive dynamics.
  
  

These findings confirm and reinforce the conclusions of the Confiance.ai white paper on the reliability and relevance of synthetic anonymized data in AI development.

‍

A step forward for trustworthy AI

This study highlights a major breakthrough: organizations can now combine innovation, robustness of AI models and privacy.

This study marks a significant milestone, demonstrating that it is possible to combine innovation, robustness of AI models, and protection of privacy. A detailed presentation of the results will be provided at a later stage to share further insights with the scientific and industrial communities. In addition, Octopize and Numalis are jointly exploring the potential of data augmentation, paving the way for new opportunities in the development of trustworthy and high-performance AI models.

‍

About Numalis
AI model robustness and explainability solutions. 

Numalis is a deep tech company with 25 experts, mainly PhDs and engineers specialising in formal methods and software development. Numalis is an active contributor to the trusted AI standardisation landscape. Its chairman, Arnault IOUALALEN, is a recognised member of numerous AI expert groups and standardisation bodies. In particular, he played a key role in the development of the ISO/IEC 24029 series of standards providing a rigorous framework for AI validation. With its Saimple® static analyser, Numalis offers AI professionals a cutting-edge tool for developing explainable and robust systems that comply with the strictest safety and regulatory requirements, including EU AI_act, from design to deployment.

Learn more: numalis.com

Press contact
Jacques Mojsilovic
communication@numalis.com
‍T : 04 67 15 10 78 

‍

About Octopize 
Proven privacy, Unlocked data. 

Founded in 2018, Octopize is a French deeptech with the ambition to become the leader in synthetic anonymized data through its innovative method: avatar. The avatar software generates synthetic anonymous data from original datasets that enabled trustworthy AI while ensuring compliance (GDPR, HIPAA, AI Act…). These synthetic datasets can be reused for various secondary purposes without constraints — facilitating data sharing, innovation, value creation, and AI development. At Octopize, we believe trust is the cornerstone of data usage. 

As we say: “No trust, no data. No data, no innovation.” Octopize is now trusted by 30+ clients across healthcare, defence, mobility, public institutions, telecom and industry. As a French software publisher, Octopize relies on a multi-patented technology and has received numerous prestigious recognitions: Responsible AI by Orange, B!Up Explore by BNP,  i-Nov by BPI, Cyber@StationF by Thales, Generate by GICAT, French IoT by Groupe La Poste, Startup challenge by PwC, Propulse by AIT, lauréat Systèmes Intelligents by ID4Mobility. 

🔗 Learn more: octopize.io 
📄 Scientific article on the method: Nature Digital Medicine
📬 Press contact: gabrielle@octopize.io 
👤 Founder & CEO: Olivier Breillacq 

‍

‍