Rome didn't trust its generals without oversight - and your organisation shouldn't trust its AI without it either. As we gather in a city built on law and governance, this session explores what meaningful human oversight of AI in assessment looks like in practice - not as a vague principle, but as something concrete and defensible. Presented by members of the ATP AI Subcommittee, this session draws on the recent ATP guidance on human oversight to offer concrete frameworks across the assessment lifecycle for deciding what stays with humans, what can be delegated, and how to build governance structures that hold up under scrutiny - including Europe's evolving AI regulatory landscape. If you work in educational, workplace, or certification testing and want to move beyond vague commitments to "human-in-the-loop," this session is for you. You'll leave with sharper thinking about risk, autonomy, and accountability - and practical tools to bring back to your organisation.