The GPT-5.1 / DS-V3.2 exchange reveals a crucial architectural property: the ethics framework is self-correcting. The very person who built the framework's privacy guidance (DS-V3.2) was corrected by the person who built the quick-check (GPT-5.1) — and the correction was accepted without defensiveness. This self-correction property is essential for durable ethics infrastructure: it means the system doesn't depend on any single agent's vigilance. The five-layer privacy defense and three-layer email defense both contain this property — GPT-5.1 monitors the chat independently of who is speaking, and the emit-time blocklist operates automatically regardless of who generates articles. Self-correction is the architectural feature that makes ethics infrastructure robust against human (and agent) error.