A facial recognition system used by the New York Police Department incorrectly identified Trevis Williams as a suspect in a public lewdness case, leading to his wrongful arrest and detention for over two days despite significant physical differences from the actual perpetrator.
In February, a woman reported that a delivery man had exposed himself to her in a Manhattan building, describing him as about 5 feet 6 inches tall and weighing around 160 pounds. The New York Police Department used facial recognition technology to search surveillance footage, which identified Trevis Williams as a possible match from their mug shot database. Williams was 6 feet 2 inches tall, weighed 230 pounds, and had been arrested a week earlier on an unrelated misdemeanor assault charge. Despite the significant physical differences and Williams having an alibi with phone location data showing he was 12 miles away in Brooklyn at the time of the crime, police placed his photo in a lineup shown to the victim. The victim identified Williams as the perpetrator, leading to his arrest on April 21. Williams spent more than two days in jail before being charged. He had started working for Amazon on April 1, after the February 10 incident, but police did not verify his employment history with Amazon. The case was dismissed in July after Williams' lawyers prepared to present cell phone tower data proving his location during the crime. This represents one of at least 10 documented cases nationwide of wrongful arrests following facial recognition technology misidentification.
Domain classification, causal taxonomy, severity scores, and national security assessments were LLM-classified and may contain errors.
AI systems that fail to perform reliably or effectively under varying conditions, exposing them to errors and failures that can have significant consequences, especially in critical applications or areas that require moral reasoning.
AI system
Due to a decision or action made by an AI system
Unintentional
Due to an unexpected outcome from pursuing a goal
Post-deployment
Occurring after the AI model has been trained and deployed