OpenAI proposed a US-led standards effort covering recursive self-improvement.

The September 21 post asks Washington to coordinate national safety institutes on measuring capabilities, testing safeguards, overseeing recursive self-improvement and reporting incidents. It names no governments, gives no timetable, sets no penalties, and leaves enforcement national. It gives Washington a lab-backed alternative to a UN body in the same week as the oversight call. OpenAI separately published four priorities and seven principles for third-party assessments, including pre-registered claims, disclosed assessor conflicts of interest, and a remediation window before publication. Voluntary, and no auditors named.