Or Hiltch

“Co-developed Boundary-Bench benchmark testing agents under realistic enterprise constraints like EDR, SASE, and DLP with researchers from NYU.”

“Co-founder who explained how the privilege escalation exploit chain worked in the Claude Cowork sandbox escape.”

“Ran 480 Claude Code workflows through hardened environments and found Fable/Mythos was the calmest behaviorally with the flattest step curve, peaking around $0.74 in the hardest setting.”