This is completely expected. First of all, Jason has interstitials in many of his stories that say things like this:

> Do you know anything else about Flock? We would love to hear from you. Using a non-work device, you can message Joseph securely on Signal at joseph.404, or Jason on jason.404

So obviously every 404 article in the training data, or which surfaces in a search, plants the seed "Jason, a reporter at 404 Media, is very interested in unethical or unfair practices relating to tech companies!

And when you tell ChatGPT "Meta/Apple/Google/PayPal is deliberately and systematically screwing me - it's unfair, unethical, and I'm pretty sure it's illegal!!" this idea is brought into the context window verbatim and taken seriously. I don't think these chatbot harnesses really do much[1] to distinguish "what user just said" from "reality." So, when people say they've discovered the corporate crime of the century, of course you should email Jason at 404 Media! Surely he would be devastated if someone else got the scoop first.

[1] I assume the only exception where it might push back on what you say might be in areas where the model has been deliberately steered away from "harm." But I don't think this type of thing would trigger that.