Notes on trustworthy enterprise AI

Engineering write-ups from the team building qbrin. Every number we publish here is measured, and we show our work.

0 invented answers across 120 trap questions: how we measure qbrin's trust layer

July 13, 2026 · We built adversarial question sets on four different corpora, tried to bait qbrin into making things up, and audited every single answer against the source text. Here is exactly what we found.