When Paul Vann's RSAC 2026 Conference session Beyond the Black Box: How to Detect AI Bias Before It's Too Late wrapped in March, it left the industry with a hard question: as models get more opaque, how do you catch bias before it does damage? Months later, the landscape has shifted, and not in the direction anyone hoped.

In this fireside chat, Paul revisits the original argument in light of what's changed: why AI bias detection has gotten harder, and whether the interpretability research he pointed to as a way forward has actually gone anywhere.