The Quick Version
- At least 13 criminal cases have been dismissed after facial recognition misidentified the wrong person, and nearly every known victim has been Black.
- Cases like Robert Williams and Porcha Woodruff show the human cost of treating a software match as proof rather than a lead.
- Federal testing finds many systems misidentify Black faces at far higher rates, compounding over-policing in Black communities.
- More than 20 jurisdictions have restricted the technology, and none with an active ban has reported a wrongful arrest.
Imagine being pulled from your driveway in front of your children, arrested for a crime committed by someone you have never met, in a city you may have never visited. For a growing list of Black Americans, that is not a hypothetical. It is a police report generated by a computer.
A Pattern Too Consistent to Ignore
At least 13 criminal cases across the country have now been dismissed after facial recognition technology matched the wrong person. What should stop you cold is the demographic detail: nearly every known victim of these misidentifications has been Black. The Innocence Project has been tracking how these tools put innocent people at risk.
The names are worth saying. Robert Williams, arrested in front of his family in Detroit after software misidentified him. Porcha Woodruff, eight months pregnant, detained on a carjacking charge she had nothing to do with. Nijeer Parks, jailed for days over a false match. Randal Reid, arrested in one state for a crime in another he had never entered. Each case settled or dismissed. None of that time comes back.
The Machine Has a Blind Spot — and It Is Us
This is not only a story about lazy policing. It is a story about math. Federal testing has repeatedly found that many facial recognition systems misidentify Black and Asian faces at far higher rates than white faces. Build a tool on training data that underrepresents darker skin, deploy it on communities that are already over-policed, and you get exactly the pattern we are seeing.
Here is where I try to be fair. Facial recognition can genuinely help investigators, and no serious person wants unsolved crimes. But a lead is not probable cause. Too many departments have treated a software “match” as a conclusion rather than a starting point, skipping the basic corroboration that would have cleared these people in an afternoon.
What Guardrails Actually Look Like
More than 20 jurisdictions, including Boston and San Francisco, have restricted or banned police use of the technology, and reformers note that no wrongful arrest has been reported in a city with an active ban. That is not a coincidence. Where the tool is fenced in, the harm follows.
What You Can Do Before It Reaches Your Door
- Ask how your city uses it. Public records requests and city-council questions force departments to disclose whether facial recognition is in play.
- Push for a corroboration rule. Demand written policies that bar arrests based on a match alone.
- Know your rights during a stop. You can ask what evidence connects you to a crime and can decline to answer questions without a lawyer.
- Support local transparency ordinances. The strongest limits have come from city councils, not Congress.
For more perspective on technology and civil rights, visit our Voices & Perspectives archive.
Technology is supposed to reduce human error, not launder it. When a system fails in the same direction every single time — against the same faces, in the same neighborhoods — that is not a glitch. It is a design we are choosing to keep. We can choose differently, and the cities that already have are showing the rest of us how.
