Safeguarded AI
About
Updated 05/18/26Safeguarded AI is ARIA’s flagship AI safety programme within the Mathematics for Safe AI opportunity space. Backed by £59m over four years, it funds research to combine scientific world models with formal mathematical proofs so that AI systems operating critical infrastructure can come with quantitative safety guarantees, analogous to certification in nuclear power or passenger aviation. The programme is structured into three technical areas that develop formal platforms, use frontier AI to help experts build verifiable models of real‑world systems, and demonstrate the resulting assurance tools in safety‑critical applications, with a growing focus on cybersecurity and microelectronics.
Theory of Change
Safeguarded AI assumes that empirical testing alone cannot reliably guarantee the behaviour of highly capable AI systems in safety‑critical settings. Its theory of change is that by combining scientific world models with formal proofs, researchers can build gatekeeper AIs and a general mathematical assurance toolkit that make AI behaviour independently verifiable. If this infrastructure is adopted across sectors such as energy, healthcare, and cyber‑physical infrastructure, organisations will be able to deploy advanced AI with quantifiable risk bounds, reducing the chance of catastrophic failures while enabling large efficiency and performance gains.
Community Signal
Updated 05/18/26Endorsements support Advanced Research + Invention Agency (ARIA).
No endorsements yet.
Discussion
No comments yet. Be the first to share your thoughts.
Details
- Start Date
- Nov 12, 2024
- End Date
- -
- Expected Duration
- -
- Funding Raised to Date
- $59,000,000