His technical governance research applies systems-theoretic hazard analysis to loss-of-control risk in the internal deployment of frontier AI systems, and he has contributed to work on verifiable compute auditing: protocols by which a datacentre can demonstrate compliance with constraints on frontier training without disclosing confidential information. His alignment interests are mathematical, centring on world models and transducers, their homomorphisms and mereology, and on the algebro-geometric foundations of singular learning theory.
Aurelio trained as an algebraic geometer, completing a DPhil at Oxford under Balázs Szendrői and holding a JSPS Research Fellowship at the Kavli IPMU, University of Tokyo, before turning to AI safety in 2026. He has since completed research fellowships with SPAR and Dovetail Research.