LLMs are good at many things, but subtlety (and ability to slowly do anything consistently over time) is not remotely one of them. I don't know why people keep bringing this idea up - if they were prompted to do something, you would notice. Don't you remember how noticeable it was when chatgpt gave follow up questions for engagement on EVERY damn response? It's got the finesse of a bulldozer.

I encourage people thinking about such things to try and build prompts to emulate this behavior and see how well it does.