Our CEO at UNESCO's 4th Global Forum on the Ethics of AI
Eticas's Founder & CEO, Dr. Gemma Galdon-Clavell, is designing and leading a workshop for the Global Network of AI Supervisory Authorities (GNAIS) at the 4th UNESCO Global Forum on the Ethics of AI, taking place in Riyadh from 14 to 17 September 2026. Convened by UNESCO, hosted by Saudi Arabia's SDAIA, and co-organized with ICAIRE, the Forum brings together 194 member states and more than 1,200 policymakers, regulators and researchers under the theme "Transforming Global Cooperation for Ethical AI Governance." Gemma will also speak on the panel "Synergies between AI Safety and Ethics" and join the Day 3 plenary, "AI for the People: Towards a Shared Future."
What GNAIS is built to solve
GNAIS - the Global Network of AI Supervisory Authorities - launched at the 2025 Global Forum in Bangkok as a UNESCO platform for the government bodies now responsible for overseeing AI in their countries. It grew out of an informal working group of European AI regulators, which UNESCO scaled into a global network chaired by Saudi Arabia's SDAIA. Its members span very different mandates, legal powers and technical capacity, but they share the same new problem: supervising systems that, until recently, were nobody's job to check.
The gap between policy and evaluation
Most of the frameworks a supervisory authority inherits - national AI strategies, the UNESCO Recommendation on the Ethics of AI, sector-specific guidelines - describe what an AI system should do: be fair, transparent, safe, accountable. Far fewer specify how a regulator, often with limited technical staff, actually checks whether a deployed system meets that bar in practice, rather than on paper.
That's the same gap we work on every day at Eticas.ai, evaluating high-risk AI systems - expert and predictive models, LLM-based tools, agentic systems - in production for public- and private-sector clients. That work has shown at scale that closing the gap doesn't require every regulator to become an AI lab. It requires a shared, testable methodology: a defined way of turning a principle like "fairness" into a measurable check that produces the same verdict regardless of who runs it.
What a structured evaluation actually looks like
That's the substance of the workshop Gemma is leading for GNAIS: what does independent AI evaluation look like in practice, and what can supervisory authorities standardize across borders so they aren't each reinventing it from scratch?
Some of that thinking is already public. Our AI Risk Taxonomy lays out a four-layer method for turning a stated risk into a mechanism, a technical probe, a metric and a graded outcome - tested against 18 existing frameworks, including the EU AI Act and NIST. Regulators face the same translation problem organizations deploying AI do. The difference is that when a regulator gets it wrong, an entire population inherits the risk, not just one company's customers.
Two things follow from that, and both are relevant to how GNAIS members are thinking about supervision. First, evaluation has to happen where the system actually runs, not in a lab or on a vendor's benchmark - a model that passes a pre-deployment test can still drift once it meets real applicants, real patients or real claimants. Second, a one-time check is not oversight. Systems change, populations change, and the risk profile a regulator signed off on a year ago may no longer hold. That is why, in our own work with clients, evaluation is always followed by continuous monitoring against the benchmarks it establishes - and it is the same logic supervisory authorities are now applying to entire markets, not single deployments.
Where this goes next
GNAIS is little more than a year old, and the practical machinery of AI supervision - who tests what, with what authority, against what evidence - is still being built in real time, in rooms like the one in Riyadh this week. That is exactly why it is worth being in the room: the standards regulators adopt now for evaluating AI will shape what "compliant" actually means for every organization deploying AI, for years to come.
If your organization needs to demonstrate that its AI systems hold up to the kind of scrutiny regulators are now building the capacity to apply, let's talk.