And the guardrails possible on AI are laughably bad. For fun, look up some of the leaked system prompts for stuff like ChatGPT. They're just using the same sort of "ask it nicely to do what you want" as the end users are, they just re-send it before almost every command.
The other big option is keyword based filtering, and anyone who's played any MMO in the last 30 years can tell you how effective that is. People never find new ways around keyword filters.
Not to be rude, but that shouldn't be news, and it sure as hell isn't something new specifically under Trump or the Republicans. Talk to some old heads, or read what they've written. US cops have stalked people for less, and there's plenty of declassified (and leaked if you're not afraid of being put on a list) guides out there from US intelligence agencies and their numerous contracted training groups that also train cops on how to infiltrate grass roots movements, rallies, etc.
This is what a chunk of these people literally do for a living.