Plaid is the main risk - if they were hacked, that's way worse than a log of my AI queries. What's exciting with LLMs is the possibility to do classification much better.