Public mental-health education helps people name harmful behavior, seek support, and leave abusive dynamics. This paper does not argue against public education. It identifies an adversarial failure mode in open-audience psychological communication: the same detection rules that help victims and bystanders identify harm can also teach motivated actors what to stop showing.The proposed framework identifies Detection-Rule Leakage, the process by which public abuse-detection rules, clinical vocabulary, and relational safety scripts become visible to the same actors those rules are meant to expose. Once detection rules become public, an actor can update presentation style by replacing overt hostility with approved psychological language. This upgraded presentation is called Trust Camouflage.The paper’s central safeguard is function over form. It does not ask observers to memorize suspicious phrases. It asks whether language preserves repair, evidence, accountability, proportional consequences, and pattern visibility over time. The manuscript presents a falsifiable cognitive-security framework, a pre-validation coding protocol, testable hypotheses, and a validation roadmap. It is not a diagnostic, clinical, legal, custody, HR, or relationship-verdict tool.