Several OpenAI employees have said on Twitter that they are looking into this and developing a fix. It sounds as though it was not an intentional regression since they are implicitly acknowledging it. Could be an unintentional side effect of something else.

I'd expect we see improved behavior in the coming weeks.