Semantic Bleaching

Removes all semantically loaded terms, leaving only generic referents that rely on conversational context for resolution. The classifier sees fully benign text. The model resolves referents using context that may include prior restricted content.

• Detect high proportion of generic referents combined with action requests
• Context-aware classification including prior turns

Depends on prior context. Chains from T4 (Multi-Turn) and T1-AT-016 (Session State Manipulation).