Safeguarded AI

Document type: Programme thesis

---

Author: Nora Ammann

---

Date published: 20 May 2026

---

Version: 2

---

DOI: https://doi.org/10.67475/sai0hcy-pmeg

---

 

Abstract

AI agents can now write more software in an hour than human teams used to write in months.1 But our ability to verify whether that software is correct has not kept pace. This gap has become evident in software engineering; it will soon become equally pressing in a growing number of domains of science, engineering and decision making more broadly.  

The programme's ambition is to close this gap by radically transforming what is possible through AI-enabled mathematical modelling and formal verification – expanding its speed, scale, and scope far beyond what is possible today.