I really doubt it’s OpenAI’s problem in regards to Machine Learning. Or a bug.
They most likely set a threshold and if you fall below it you are given a captcha.
This would lead me to believe that one of your plugins is doing things that triggers anti-bot security. Using any privacy plugins like PrivacyBadger by any chance? It goes both ways. Why only for the moment? No idea.
With some captcha systems like Google ReCaptcha your score is universal. As in, the same score on PageX travels to PageY. It’s kind of creepy. So even changing browsers can trigger a captcha
And yes, I did read that you don’t have these issues anymore. It could be worth discussing regardless so other people can read and potentially find commonalities.
One of the common ways to check a bot is by seeing if they always enter a page with fresh settings (As testing frameworks typically discard all settings to guarantee the same run-through on the next testing run)