What Did Day 2 Reveal About AI Ethics, Oversight, and Risk?
Sep 26, 2025
We’re back with Part 2 of our Eyes-Off Data Summit 2025 roundup. Here are the resources and topics discussed at the event, covering law and AI ethics, red teaming, mosaic memory, and global AI compliance.
One Article
One of the most insightful talks at the Summit came from Valerie Lyons, who dissected the gap between law and ethics in AI. In this article, she calls for clearer enforcement, addresses gaps in current regulation, and urges a sharper conversation about what real responsibility in AI governance should look like.
One Infographic
In his day 2 presentation, Martin Mannion (Deloitte) spoke about AI compliance engineering and embedding trust in AI delivery, where he spotlighted the complexity of global AI compliance.
Source: modulos.ai
One Technique
Liubov Nedoshivina (IBM) talked about red teaming, which has become a vital technique for exposing the hidden risks in generative AI. This deep dive by IBM explains how researchers use adversarial prompts, synthetic troll models, and automated tools to identify vulnerabilities.
One Research Paper
This paper, co-authored by Georgi Ganev (SAS), reveals that how you discretise data, the step where continuous values are grouped into bins, can impact both privacy and performance in DP models. Done right, it boosts utility by 30% and mitigates the risk of membership inference attacks.
One Album
Here’s the official photo album from the Eyes-Off Data Summit 2025 in Dublin that happened last week. Browse the best moments or imagine yourself there if you couldn’t make it. We’ll be posting more talk recordings, photo highlights, speaker insights, and blog reflections on LinkedIn.
