DevFest Montréal 2026

← Back to the programme
3-5 · Just Can’t Get EnoughEnglish

Beyond Red, Blue and Purple: Meet the Gold Team

Category
AI & ML
Format
Session — 40 minutes
Audience level
All levels

Red teams ask: Can I break it?
 Blue teams ask: Can I detect and stop it?
 Purple teams ask: Can Red and Blue learn from each other? But there is another question we rarely ask: Why did it break in the first place? That is the job of a Gold Team: a Governance-Oriented LLM Defense team dedicated to taking AI governance to another level of understanding. Because you cannot govern what you do not understand. During this talk, Faustin Bouchard will introduce the first Gold approach to understanding LLM failure: reconstructing the logical steps an LLM goes through to reach a decision, then examining how and why each of those steps can fail when the system encounters an adversarial prompt. And, surprisingly, once you see the structure, it becomes remarkably simple.

Speakers

Faustin Bouchard

System Architect · Universal Intelligence Architecture Inc.

Faustin Bouchard is an independent AI safety researcher and systems architect focused on adversarial LLM evaluation, black-box system identification, behavioral failure analysis and runtime governance. His path into AI research began with a long-term study of recurrent behavior under constraint. As high-performing LLMs became widely available, he translated those observations into computational hypotheses and began testing them experimentally through controlled perturbation, cross-model evaluation, and adversarial stress testing. Over the past 14 months, he has led the development of the Universal Intelligence Architecture™, spanning computational architecture, experimental design, adversarial corpus development, model evaluation and governance research.