Which OpenAI sessions can misalignment monitoring stop?
OpenAI's misalignment monitoring can automatically stop only Responses API conversations that preserve context through persisted reasoning, WebSocket continuation, or server-side compaction. Other Responses requests are still monitored but only emit alerts, and Chat Completions sits outside the system. Treat it as a backstop. It is not a permission control.
4 min read