Could AI Face Recognition Falsely ID You?

The police have your picture. Maybe this doesn’t worry you. But every day, the police are comparing your picture to surveillance pictures of criminals. And if you share facial features with actual culprits, you could be the next person wrongly investigated and arrested for a crime. Maybe you should be worried.

Similarity of Facial Features

“Do you know you look like so and so?” This is a common question. We all share facial features with many other people. Maybe you’ve even met your own doppelganger, someone who looks almost exactly like you. Unfortunately, this basic similarity among us puts innocent people at risk of wrongful arrest.

Cameras surround us, and we are at risk. Cameras are on street corners and in stores. You might even have a camera at your door, allowing you to check who is knocking. We also carry a camera in our pocket, take pictures frequently, and often post those pictures online.

The government also has pictures of you. At minimum, the government has your driver’s license photo and possibly a passport photo. Those pictures are in large databases used by police in facial-recognition technology systems. You are in those systems, and so am I.

Facial Recognition Technology Puts You at Risk

How do police develop suspects in a crime? Police interview witnesses and victims. They get basic descriptions of the culprit. Using that information, police investigators contact other people in the area who may have seen or know something about the crime. Law enforcement adds crime scene fingerprints or blood to a DNA database; you’ve seen this on TV shows.

Police also frequently feed surveillance camera images into facial recognition software (Bergold and Kovera, 2024). Police start with a still image from a camera that was near the crime or accident. Many are poor-quality images. They are often taken from a distance, a bad angle, or in poor lighting. Some may have been taken using fisheye lenses that distort the shape of someone’s face.

Facial recognition programs have millions of pictures. Those pictures come from everywhere in the country. Those photos also vary in age. How old is your driver’s license photo, and do you still look like that? How many photos of you exist on social media?

Using facial recognition tech, police will get hits. The police may then use images and other database information to start an investigation and possibly make an arrest.

Perhaps this all sounds reasonable. But the systems search through millions of photos from all over. We all share features with lots of other people, including some who have committed crimes. If the facial recognition tech says you look like the culprit, the police may create a photo lineup with your picture to show the witness or victim. Anyone added to a lineup based on these systems will likely look like the culprit; this is why the system tagged them in the first place. Witnesses can certainly identify an innocent person (Lindsay and colleagues, 2025).

How Bad Is Your Risk?

Assessing risk depends on knowing the false alarm rate—how often this tech misidentifies an innocent person as the culprit. We don’t have clear information on how often the systems make false alarms. But we do know some things that should concern everyone. Bergold and Kovera (2024) noted that some systems, such as the one used in New York City, will return 200 possibilities when given a starting image. At best, 199 are false alarms, innocent people who happen to resemble the person in the starting picture. That’s a 99.5 percent false alarm rate. And it could easily be that all 200 are false alarms, not the person in the image, a 100 percent false alarm rate. That’s a lot of false alarms, a lot of innocent people at risk of investigation and false charges. And the bigger the photo base, the more false alarms. If police try to investigate all 200 people returned, that’s a huge drain on resources. You may be one of the innocent people who is wrongly investigated.

The Wrongful Arrests So Far

We’re already hearing horror stories with this new technology. There’s the story of Porcha Woodruff, who was 8 months pregnant when she was wrongly arrested. The culprit was not described as pregnant. But her old driver’s license photo was matched to a surveillance photo of the culprit’s face. The police arrested and held her overnight. After more than a month and multiple hearings, law enforcement dropped the charges (Bergold and Kovera, 2024). Or Angela Lipps, the grandmother in Tennessee who was arrested for a crime in North Dakota, a state she had never visited. She was extradited to North Dakota. Again, it took months for law enforcement to drop the charges (Sottile, 2026).

The Risks for You, Me, and Everyone Else

The risks can be summarized. All of us have pictures in the facial recognition databases. We all have facial similarities with many other people, including some people who commit crimes. Most suggestions from the systems are false alarms of innocent people, and the reasons for investigating some but not others are unclear. If your face is similar to a culprit's, witnesses would pick you in a lineup. And you won’t always be told the police considered you a suspect based on facial recognition technology. We are all at risk of being wrongly arrested and prosecuted.

References