Notes from the team
Building auditable compliance AI
What we're learning making AI output gradable: evidence, citations, and the loop that argues back.
AICompliance
Silent success is the failure mode
Language models are trained to complete, not to push back. Unchecked completion, not hallucination, is the real compliance risk. On the planner → generator ⇄ evaluator loop that makes AI output gradable.
More on the way
We publish as we build. New posts land here as the team writes them.
About Tracfox
Tracfox runs your compliance program with auditable AI agents. Every step traces to a citation; every citation to a version of the law.
About Tracfox →The team
Built by people who have shipped regulated financial software and sat on both sides of an examination.
Meet the team →