I don’t know what a person “coming up with the same sentence as an llm” would have anything to do with this unless the LLM detection is based on direct string comparison.

The use of bots is inherently deceptive

Nope. I can say:

Here’s what GPT-4 generated when I gave it that prompt: “[some LLM output that would get them banned by the machine we’re proposing to build]”

That is not deceptive. But it would be detected by this system and result in them being banned. Because you guys are gung-ho to build a powerful head-cracking machine and didn’t think of an obvious edge case.