Flagship Programme
Privacy-preserving technologies for agentic AI
Private AI models and secure agentic workflows with built-in regulatory compliance.
Why this programme
The challenge we are addressing
As the quantity of data in our digital world explodes, concerns grow over fair use of personal information, secure data-sharing between organisations, and data sovereignty across borders. Advances in AI create powerful incentives to extract value from these vast information stores, but this value is locked behind regulatory requirements.
Safe navigation of these requirements needs careful application of privacy-preserving technologies. Existing approaches such as anonymisation can destroy utility without guaranteeing safety. Differential privacy offers formal guarantees, but current training methods can be computationally prohibitive at scale. Secure methods for private model training promise no defence for data passing through at inference time or during inter-agent communication or tool use. Making modern digital ecosystems private requires rethinking their components, redesigning each to be natively private and fully compliant.
Our vision
We envision a world where organisations can unlock the value of data and AI without compromising individual privacy, commercial confidentiality or data sovereignty – enabled by digital systems that are private by design and provide demonstrable, end-to-end guarantees, transforming privacy from a constraint to a foundational capability of modern AI systems.
Our distinctive position
As a lab embedded in a global insurance ecosystem, we benefit from strong understanding of regulatory and risk requirements, and insight into real industry challenges where privacy technologies can unlock new value. Combined with deep AI expertise and strong ties to Oxford’s leading machine learning researchers, this enables us to translate theory into practical solutions: facilitating secure collaboration across insurers, reinsurers and financial partners; enabling privacy-preserving fraud intelligence across organisations; and deploying AI agents safely within and across organisational boundaries. By connecting academic research with engineering and commercial application, we can pursue fundamental advances while testing their practicality against real industry needs.
Team
The people leading and supporting this programme.