San Jose Piloted an AI to Scan for Homeless Encampments. The Accuracy Was 10 Percent.

What happened

In March 2024, the city of San Jose, California revealed it had been piloting car-mounted cameras paired with AI software to scan city streets for homeless encampments, vehicles that people were living in, graffiti, and trash. The system was built by SenSen AI and Zyrex under a city contract. Officials described the goal in humanitarian terms: catch small encampments early and dispatch outreach workers before conditions deteriorated.

Accuracy testing complicated that pitch. City documents showed the system correctly identified lived-in cars only 10 to 15 percent of the time. RV detection was better, at 70 to 75 percent, but that still meant roughly one in four occupied RVs was missed or misclassified. A tool that misfires on nine out of ten car identifications is not generating leads for outreach workers. It is generating noise, except the noise is attached to real people in precarious housing.

Privacy advocates and civil rights groups did not object to the outreach framing in principle. They objected to what the data could do once it existed. A camera flag marking a specific vehicle enters a city database, and from there it can travel to code enforcement, parking enforcement, or any city department with an interest in clearing the location. The people flagged have no way to know they were flagged, no way to contest the classification, and no recourse if the AI was wrong, which, for lived-in cars, it was nearly always wrong.

San Jose's communications emphasized what outreach workers would do with the information. They did not address what enforcement branches could do with the same data, or whether any policy existed to prevent it from migrating to a harsher use. That gap is the center of the civil rights complaint. The technology does not care about the city's stated intent. It produces location-tagged records of where homeless people are, available to anyone with database access.

This is the structural problem with deploying surveillance tools on a population that has limited standing to push back. No part of the pipeline, from camera to database to city worker, requires the city to document what classification was made, which official acted on it, or what happened to the person at the flagged location. A provable record of what a system did, for whom it was queried, and what response it triggered, is exactly the audit trail that would let courts and advocates test whether the compassionate framing held up in practice. Without it, the city can describe any outcome in the best possible terms, and nobody has the data to say otherwise.

Reported impact

Affected parties

Not publicly disclosed

Harm type

Not publicly disclosed

Scale

Not publicly disclosed

Financial impact

Not publicly disclosed

Regulatory action

Not publicly disclosed

Classification

Organization

Not publicly disclosed

AI system

Not publicly disclosed

Industry

Not publicly disclosed

Country

Not publicly disclosed

Provider

Not publicly disclosed

Incident type

Not publicly disclosed

Relevant governance controls

Governance control mapping is not available for this record.

• No controls mappedNot publicly disclosed

Control mapping is analytical. It does not state that any control would have prevented the incident.

Sources and evidence

This record was researched and written by the Index. The event is also catalogued in the following database, which is listed for cross-reference.