top of page
19 October 2026
|
Partnerships

Aioi Nissay Dowa Insurance Company - Japan and OXFIDA join the Japanese Government funded CODAS program to develop privacy preserving Agentic AI

​

Introduction

Aioi Nissay Dowa Insurance Co., Ltd., a member of the MS&AD Insurance Group, and OXFIDA are pleased to announce their participation in the New Energy and Industrial Technology Development Organisation (NEDO) commissioned project ‘Strengthening of Post-5G Information and Communications System Infrastructure / Development of Efficient AI Training Methods Considering the Confidentiality of Data’ (‘The Project’) as a partner to the Co-creation Organization for Data Sovereignty in AI (CODAS), established in October 2025.


OXFIDA will deliver a core R&D program within the Project advancing new technologies and methodologies supporting the development and deployment of privacy-preserving Agentic AI solutions. With a combination of Open Source Software delivery, exploratory methodological work, and business case development, grounded from an internal viewpoint of the Insurance Industry, Aioi Nissay Dowa Insurance and OXFIDA aim to become long-term partners to CODAS in their mission to deliver open, sovereign, private and secure AI tools to the world that enable businesses to confidently and safely make the most of their data.



The problem we are solving

Agentic AI has been taking the world by storm. We are no longer asking each other ‘have you used AI?’ Instead, we are now all asking ourselves, how can we get the best from AI and stay ahead of our competition?


Agentic AI is fundamentally different from traditional software. The core of Agentic AI systems are GenAI models, typically Large Language Models (LLMs). In contrast to traditional software, these models interact with tooling and with each other through arbitrary text input and output. The innovation that transformed Gen AI into Agentic AI was the invention of a software architecture that allows these models to interact with each other and with traditional software tools to perform complex tasks in a computer autonomously.

 

The deployment of Agentic AI systems comes with large data governance challenges for businesses. For example:

  • Today AI systems can leak customer personal information

  • Sensitive data may be forwarded inappropriately without consent

  • Data can be memorised through model training, and extracted through attacks on the model

  • Agentic systems do not have standardised audit trails making it difficult to create evidence of compliance in deployed systems


These risks multiply not just linearly with the number of Agentic AI systems deployed by businesses, but also combinatorially with the complexity of the Agentic systems themselves. Every new model or interface within an Agentic AI system is a new weakness or attack surface where a data breach could occur.



What we want is a system where:

  • Personal and sensitive information detected and masked in real time

  • Data anonymised before any external sharing, or external sharing is blocked if inappropriate

  • There is zero inappropriate persistence of private data through model training processes

  • Full audit trails, DPO approved, and compliance evidencing becomes possible


Our Approach

To solve this problem OXFIDA will build protections directly into the model layer of an agentic system. There are two places where technology needs to be developed:

  • At training time we need tools that make it possible to use training data for pre-training, post-training or fine tuning in a way that protects individuals represented in the training data

  • For inference time protection we need methodologies that allow models themselves to be taught how to handle personal and sensitive information appropriately when acting


With protections in place for both training data and data passing through the models at inference time, we will solve the privacy problem for Agentic Systems at its root, and avoid the combinatorial explosion in data breach risk.

 

[General Incorporated Association CODAS (Co-creation Organization for Data Sovereignty in AI)]

CODAS (Co-creation Organization for Data Sovereignty in AI) is a non-profit, neutral general incorporated association that promotes the research and development of data privacy technologies and their social implementation through open-source deployment, aiming to realise a "safe and secure AI society."

Currently, while many companies are advancing their use of AI, they face challenges such as the risk of confidential information leakage and the loss of data sovereignty. CODAS addresses these challenges by building a "Privacy-First AI" infrastructure that enables advanced AI utilisation without data ever leaving the organisation.


CODAS website: https://codas-ai.org/jp

E-mail: info@codas-ai.org

Press Release: CODAS Launch Event (Japanese only)


[OXFIDA]

OXFIDA is an Oxford-based AI and technology company established in 2022 as part of Aioi Nissay Dowa Insurance.

OXFIDA combines AI technology and commercialisation capabilities with collaboration across Oxford’s ecosystem, using large-scale accident and risk data to develop trusted AI-powered insurance and risk solutions, while researching frontier technologies including privacy-preserving AI and quantum optimisation.

Through further development of privacy-preserving technologies, OXFIDA aims to strengthen Sovereign AI, one of its key strategic pillars, and build trust in the secure deployment of advanced AI.


OXFIDA website: https://www.oxfida.ai


Delivering Community Tools and Value for Business 

Through CODAS, OXFIDA will publish its technical results and developed tools in Open Source code and published academic work. Together with its partners with the MS&AD Insurance Group, OXFIDA will also work to develop, detail and evidence the business case and value for the tools and methodologies it is creating. As an AI and technology company, with its original roots from the University of Oxford AI ecosystem, sitting in the heart of one of the largest insurance companies in the world, we are in a unique position to develop and demonstrate the value of CODAS’s vision for a world where AI tools enable sovereign control over their data, and safe use of it for their customers.

bottom of page