She pointed to training gaps.
Recordkeeping.
Language access.
Accessibility.
Data.
One state senator joked:
“You never celebrate.”
“I celebrated for two hours.”
“What more do you want?”
“Function.”
Marcus again.
Maintenance.
During this period, Westridge faced a new kind of problem.
Not a Patricia problem.
A technology problem.
The HOA upgraded access and security system.
Residents received an app.
Guest passes.
Gate codes.
Delivery notifications.
License-plate recognition at vehicle entrance.
Optional camera alerts.
The vendor, SafeGate Residential Systems, marketed an “anomaly detection” feature.
It used patterns from access logs and resident reports to flag unusual activity.
The board did not describe it as artificial intelligence.
The sales team used phrase “smart neighborhood awareness.”
Nia heard and immediately distrusted it.
Her daughter Camille, now twenty-nine and working as an investigative reporter, distrusted it more.
“You bought an algorithm.”
Nia sighed.
“I did not buy anything.”
“Your old HOA.”
“I haven’t lived there in thirty years.”
“Still yours emotionally.”
“Therapy language is abuse.”
Camille laughed.
The first months were uneventful.
Then Evan Brooks called Nia.
“Remember when you said our house would become ordinary?”
Nia’s stomach tightened.
“What happened?”
“My son keeps getting flagged.”
The nine-year-old Nia had met was now nineteen.
Malik Brooks.
College sophomore commuting from home.
“Flagged how?”
“SafeGate notifications.”
“Unknown pedestrian near private residence.”
“Unverified entry.”
“He lives there.”
Nia frowned.
“Resident profile?”
“Current.”
“Gate app?”
“Current.”
“Face recognition?”
Evan sighed.
“They say it’s not facial recognition.”
“Then what is it?”
“Behavioral matching, apparently.”
Nia called Camille.
Camille came to Westridge with notebook.
Not because family drama.
Because algorithms affecting residential security were a story.
She spent six weeks reporting.
The problem was more complicated than she expected.
SafeGate did not identify faces by name.
It used motion patterns, access events, license-plate history, resident reports, time-of-day data, and camera object classification.
If a person appeared near a property without a corresponding gate-entry event, guest pass, known vehicle, or phone Bluetooth token, system might flag.
That sounded neutral.
Except data inputs were not neutral.
Residents who walked in through pedestrian paths rather than driving were flagged more often.
Teenagers did that.
Domestic workers did that.
Delivery workers.
Older residents whose phones were off.
People using borrowed cars.
Tenants rather than owners.
And because resident-generated “suspicious activity” reports were part of training data, neighborhoods with historical reporting bias produced more flags for people resembling those who had been reported before.
Not necessarily faces.
Patterns.
Clothing.
Movement.
Time.
Location.
The system learned history without understanding history.
Camille interviewed SafeGate executives.
They insisted tool never instructed police contact.
True.
It only generated “review alerts.”
But Westridge staff had begun treating repeated alerts as evidence of suspiciousness.
Malik had been flagged eleven times in four months.
A white college student on same block who also walked home at night had been flagged twice.
Why difference?
SafeGate said data incomplete.
Camille asked for audit.