GitNation
Log in
Fabio Dias

Fabio Dias

Dr. Fabio Dias is the Programme Director of the FinTech and Policy MSc at the University of Surrey and a researcher for the SAEF Centre. A multidisciplinary academic with a background in computer science and statistics, he holds a PhD in Econometrics from University College London (UCL). Before his current academic tenure, Dr. Dias acquired extensive industry experience, holding senior roles in technology and risk management at major financial institutions, including Credit Suisse and Lloyds Banking Group. His research interests lie at the intersection of finance and technology, specifically focusing on financial modelling, artificial intelligence applications in finance, and quantitative risk management.
University of Surrey, UKfabio-dias-stalwart
Proving What an AI Won't Do
TechLead Conf London 2026: Adopting AI in Orgs EditionTechLead Conf London 2026: Adopting AI in Orgs Edition
Upcoming
Proving What an AI Won't Do
How do you know an AI system will refuse the thing it must always refuse? Today the answer is testing: ask it a thousand ways and hope the thousand-and-first isn't different. The policy in this talk has 55 variables a caller controls: that's 36 quadrillion situations! It also has a test suite of 33 unit tests, all green. This talk is about deciding the question instead of sampling it. A safety rule like "never do X without checking Y" becomes a logic circuit, and a circuit becomes arithmetic: equations whose only solutions are the situations the rule allows. Asking whether the rule can be broken turns into asking whether that system has any solution. Live: one rule proved unbreakable, with a short record of why that you can check yourself without trusting the program that found it; and two that break, each repaired and re-proved on the spot. This proves the rail, not the model.
artificial intelligence