top of page
​Status

The programme was launched on 31 July 2026 and is planned to continue through to May 2029.

Privacy-preserving AI agents — Japanese Government Sovereign AI

Summary

OXFIDA is participating with a team of up to 30 people in a 3-year programme, funded by the Japanese Government, that has been set up to tackle privacy and sovereignty barriers currently holding back the widespread successful deployment of Agentic AI. 

​The problem

Agentic AI is being widely adopted. We no longer ask 'have you used AI for your work', but instead ask 'how are you getting the best from AI in your business'.

But Agentic AI brings with it risks that have not yet been addressed effectively. Models such as LLMs have no intrinsic understanding of the sensitivity of different types of data, nor of how to operate under often regional and industry-specific data governance laws. At the same time, as US and Chinese model makers race ahead in the model market, companies in other countries are seeing a growing dependency on this non-sovereign technology.

Solution

OXFIDA is embarking on an ambitious R&D programme to solve this problem in partnership with CODAS, a new Japanese not-for-profit organisation 'founded to address privacy risks and data-concentration imbalances arising from the deployment of AI in society' (see CODAS's website).

We will investigate privacy technologies that address the problem of handling data in Agentic Systems according to the appropriate data governance laws, providing protection at training time and at inference time. In our work we will create open-source packages and models, publish new methodologies, and help define the business value and business case for the tools we are creating. Our work will help CODAS to build a thriving open-source community and deliver an open platform that businesses can rely on to create Agentic AI systems in their business that customers can trust.

bottom of page