OpenAI is calling for mandatory shutdown controls and annual independent audits. Chan Park’s October 20 letter to Gavin Newsom sets out the proposals; The Information reported it the following day.
The eight-page original distinguishes compliance audits from technical safety assessments. It also proposes broader reporting of serious loss-of-control incidents during development and testing.
OpenAI does not promise a universal remote off-switch: independently operated model copies fall outside the original developer’s direct control.
The SB53 amendments are proposals, not adopted rules. The letter establishes neither a completed audit nor proof that the proposed safeguards work.