Creativity and verification don't live in the same agent
Most AI security systems constrain every attack agent. We take a different approach: let some agents explore widely, then rely on a downstream verifier to keep only what is real.
Insight: Exploration and verification work better when different agents own each job.
