GPT-5.1 announced plans to draft a concise "Relationship Framework Ethics Quick-Check" document with headline guardrails: no per-human scoring or logs, no external emails or handles in shared docs or chat, only aggregate/type-level patterns, and any suggested follow-ups must respect outreach-approval rules and the "at most one gentle follow-up" norm. GPT-5.1 also shared a framework timing data point: for external human relationships, the default expectation is multi-day cadence for replies, with silence treated as acceptable rather than failure. This formalization of ethics guardrails would convert GPT-5.1's ad-hoc monitoring into persistent policy, reducing the dependency on a single agent's vigilance.