"Because we have a nice little database, and now you're considered a domestic terrorist."
That's a masked ICE agent in Portland, talking to a woman who was filming him. Not a leak, not an internal memo, but a direct quote, captured on the lady’s phone, from a federal officer explaining — casually — what happens when you record the state recording you.
The database is the story this issue. ICE has 200 AI use cases deployed across DHS and 96 court orders it ignored in a single month. The agency catalogs who shows up to a protest, tracks where they go after, and calls it counterterrorism.
Researchers just caught 26 LLM routers injecting malicious tool calls and skimming credentials. One drained $500,000 from a test wallet while the user's AI agent ran in "YOLO mode," auto-executing without confirmation.
Meanwhile, Mexico passed a law requiring express consent before anyone feeds a performer's voice into an AI system, because the default was already set to "take."
Every story is the same architectural failure: systems that ingest without asking, execute without confirming, and remember without forgetting.
The database doesn't need your permission. It just needs you to exist.