Ask HN: Stylography, AI and an impending privacy nightmare?

From what I've understood, we can make modern day computer systems exceedingly effective in recognizing patterns in (vast amounts of) data.

However, one of the ways this can be (ab)used is the de-anonymization of people through stylography. Since (plain)text datasets are relatively massive (in variety and density, not necessarily in size), one would assume that those systems (or similar ones) can also be used to analyze patterns within text and correlate those patterns with other pieces of text written by the same person.

I suppose one can mitigate this using AI / LLMs to rewrite the original source text (perhaps even multiple times), but wouldn't even better AI systems (in the future) be able to account for this and still be able to de-anonymize?

Are we transitioning towards a giant privacy cat & mouse game? Are we creating a real-life TrollTrace.com from South Park S20?

If my concerns written above are valid, then what potential solutions would you all suggest?