1) Surveillance needs to be reviewed. Even if reviewed by AI, eventually that reviewed work needs to be reviewed by a human if we're going to maintain the fiction / friction of "human in the loop". The "hits" will include false positives, unless the system is overtuned so that it rarely kicks an event.

1a) Review will take time / resources which could be spent on human policing, harming the community.

1b) Some jurisdictions may prefer "broken windows as policy", the notion that they can construct a "reasonable suspicion", given enough garbage (some of it outright garbage, the point being there is so much of it nobody cares; don't need to do an accurate drug test until trial, right?).

2) False surveillance hits will make it through human review and result in injury to innocent humans.

3) Police forces already lack the money / manpower to investigate potential crimes.

4) Police forces already "prioritize" other matters than the mentally ill setting their houses on fire or releasing plagues of rabbits into their neighborhoods (actual things that have happened to me!).