The project defines a non-aggressive, reviewable, schema-based defensive architecture for detecting, responding to, remembering, reviewing, and regulating unsafe or suspicious AI interaction ...