Security TrustWinning the Opir Challenge by testing the edgesI won Wordcab's OPIR Challenge by probing where semantic guardrails get thin and where the classifier and generator diverge.June 29 2026//5 min read#opir#ai security#red teamingOpen post→
Model WatchOpir made guardrails look like classifiers againOpir put safety filtering back in the small-model conversation with encoder guardrails for toxicity, jailbreaks, and policy categories.May 29 2026//2 min read#opir#classification#guardrailsOpen post→