Explorez tous les épisodes du podcast UNESCO’s Hands-On AI Supervision

Effective AI supervision requires reliable benchmarking ecosystems. Nicholas Miailhe discusses why benchmarks matter, how they should be constructed, and what regulators need to know about safety evaluations. The conversation highlights emerging international efforts to standardise safety testing and ensure comparability across models.

Speaker: Nicholas Miailhe (PRISM Eval)

Interviewer: Doaa Abu Elyounes, Programme Specialist, Ethics of AI Unit, UNESCO

Hosted on Ausha. See ausha.co/privacy-policy for more information.

This episode explores how supervisory authorities can translate high-level AI risk principles into practical, operational risk-mapping processes. Nathalie Cohen discusses evaluation frameworks, data considerations, and real-world challenges identified during the roundtable exercise, providing regulators with concrete steps for structuring risk identification and prioritisation.

Speaker: Nathalie Cohen (OECD)
Interviewer: Max Kendrick, AI Strategy Coordinator & Senior Advisor, Office of the Director General, UNESCO

Hosted on Ausha. See ausha.co/privacy-policy for more information.

Supervision cannot succeed without structured engagement with developers, deployers, and industry partners. In this episode, Huub Jannsen discusses best practices for market dialogue, transparency expectations, and collaborative mechanisms that support compliance while fostering innovation.

The episode highlights how oversight bodies can build trust and shared responsibility across the AI ecosystem.

Speaker: Huub Jannsen (RDI)
Interviewer: Dafna Feinholz, Director of the Division a.i. & Chief of Section for Bioethics and Ethics of Science and Technology, UNESCO.

Hosted on Ausha. See ausha.co/privacy-policy for more information.

As AI systems introduce new attack surfaces, cybersecurity becomes a foundational element of oversight. Carlos Antunes outlines key threat vectors, resilience strategies, and practical measures supervisory authorities can implement.

This episode gives listeners a clear roadmap for integrating cybersecurity considerations into AI supervision workflows.

Speaker: Carlos Antunes (Portugal National Cybersecurity Agency)
Interviewer: Yannic Duller, Project Consultant, Ethics of AI Unit, UNESCO

Hosted on Ausha. See ausha.co/privacy-policy for more information.

Red teaming is rapidly becoming a critical component of AI oversight. In this episode, Rumman Chowdhury explains how structured adversarial testing can uncover system vulnerabilities, model failures, and misuse pathways.

The discussion focuses on practical red-teaming approaches that supervisory authorities can adopt, even with limited resources.

Speaker: Rumman Chowdhury (Human Intelligence)
Interviewer: Mirela Kmetic-Marceau, Project Consultant, Ethics of AI Unit, UNESCO

Hosted on Ausha. See ausha.co/privacy-policy for more information.

A deep dive into the metrics and methodologies essential for robust AI evaluations. Agnès Delaborde examines measurement challenges, standards alignment, and the tools supervisory authorities need to assess AI system performance.

The conversation highlights gaps between emerging benchmarks and real-world regulatory needs.

Speaker: Agnès Delaborde (Laboratoire national de métrologie et d'essais – LNE)
Interviewer: Lihui Xu, Programme Specialist, Ethics of AI Unit, UNESCO

Hosted on Ausha. See ausha.co/privacy-policy for more information.