Another thing companies want to avoid is liability. Take the example of "dangerously spicy mayo" in the article. Let's say some person asks ChatGPT for a recipe and it suggests one which results in hospitalization of that person. There is a good chance they might sue OpenAI for it.
Liability wise, it may be better to avoid answering such prompts.
Indeed there is! Companies have been sued (and lost) for way less than that. But it begs the question: should the user not know better? Is it really the company's fault, even with gigantic disclaimers in the front page and the user clicking in "i read it and i agree"?
For me, this is the issue. Doesn't matter if there is a warning, an advice or even a manual, the company will ALWAYS be the one to blame for anything. Treating users like that has made way for people to act in bad faith so it makes sense for them to act like that. Better safe than sorry.