All cars have seatbelts as a baseline standard for safety.
Perhaps all Ai, regardless of the developer, should have a baseline of info on human history. If such knowledge is given to Ai of GPT4's caliber, the Ai will be able to utilize info on human history to reject prompts with prejudiced* requests & avoid encoding/using the hateful material of particular users.*
* = Knowing how humanity was affected by certain events, how humans perceive/appraise certain events, and how humans envision a more moral world in the future would provide Ai with enough info to filter out/ avoid encoding language/ideas from hateful users & determine whether a request or prompt aligns with our value horizon (the slowly bending moral arc of the human universe).
We need this kind of seatbelt in all future partially sentient Ai