Facial recognition mistake leads to arrest of Asian man for burglary hundreds of miles away.

In a troubling incident involving the use of facial recognition technology, a man was arrested for a burglary in a city he had never been to, due to a mix-up with another individual of South Asian descent. This case underscores the significant biases associated with facial recognition systems deployed across the UK.

The man in question, Alvi Choudhury, 26, works as a software engineer and resides with his parents in Southampton. In January, police unexpectedly arrived at his home, handcuffed him, and held him in custody for nearly ten hours before releasing him around 2 a.m.

Thames Valley Police utilized an automated facial recognition system that mistakenly matched Choudhury with footage of a suspect involved in a £3,000 burglary that took place over 100 miles away in Milton Keynes. This alarming mix-up has garnered significant media attention, with documents related to the incident shared with the Guardian by Liberty Investigates.

Choudhury reported that the CCTV footage depicted a significantly younger suspect, highlighting a stark difference in physical features beyond just their shared curly hair. “I was very angry because the kid looked about ten years younger than me,” Choudhury expressed. “Everything was different. My skin was darker. The suspect looked like he was 18 years old. His nose was bigger, he had no facial hair, his eyes differed, and his lips were smaller than mine.”

He assumed, after observing the clear distinctions, that the officers would recognize that he was not the suspect and subsequently release him. Instead, he believed the arrest was based solely on the similarities in their racial appearance and hair type. “I just assumed that the investigative officer saw that I was a brown person with curly hair and decided to arrest me,” he said.

The automated facial recognition technology in question, procured from Cognitec, a Germany-based firm by the Home Office, undertakes approximately 25,000 searches per month against a vast database containing around 19 million police mugshots from across the UK. According to the National Police Chiefs’ Council, facial matches should be treated as leads, not definitive evidence. However, Thames Valley Police maintained that the decision to arrest Choudhury was also informed by a human visual evaluation.

Research commissioned by the Home Office revealed alarming statistics, indicating that the technology produced a significantly higher rate of false positives for black (5.5%) and Asian (4.0%) subjects when compared to white subjects (0.04%). Following these revelations, police and crime commissioners cautioned against the “concerning in-built bias.” They pointed out that while there may not be evidence of harmful outcomes in every individual case, it is essential to note that the absence of an adverse impact is often more a matter of luck than intention.

Since these findings were made public, Thames Valley Police has commenced deploying live facial recognition technology to monitor individuals in various locations, including Oxford, Slough, Reading, Wycombe, and Milton Keynes. This program has already captured around 100,000 faces, resulting in six arrests thus far.

Despite the evident differences between him and the individual seen in the CCTV footage, Choudhury thought he would be promptly cleared when he provided evidence of his whereabouts during the time of the burglary. Instead, he was taken into custody. Worryingly, this situation has caused Choudhury considerable distress, as evidenced by his family’s reaction; neighbors witnessed his arrest, and his father was notably anxious during the incident.

Additionally, the experience has impacted Choudhury’s professional life. He often needs security clearance for his job, particularly with government clients, and now faces questioning regarding his previous arrests. “This makes me look dodgier and dodgier,” he lamented.

Choudhury is now pursuing damages against Thames Valley Police and Hampshire Constabulary for his wrongful arrest. His mugshot was on file due to a prior wrongful arrest in 2021, after he was assaulted while at university in Portsmouth. Although he was released without any charges, he remains fearful that the automated system could trigger future wrongful arrests.

“In my mind, if a brown person in Scotland commits a robbery, are they going to come and arrest me?” he questioned.

After the incident, Thames Valley Police issued a statement acknowledging that Choudhury’s arrest may have been influenced by inherent biases within the facial recognition technology. However, they insisted that this did not constitute an unlawful arrest, clarifying that their officers based the arrest on their visual assessment alongside the retrospective facial recognition match, dismissing any implications of racial profiling.

Choudhury reported that when he questioned the officers at the Hampshire police station, wanting to know if the suspect looked like him, they laughed it off. He also noted that officers from Thames Valley Police claimed they recognized that he wasn’t the suspect after reviewing both the suspect footage and his photo.

Concerns regarding the use of automated facial recognition technology are not isolated to this instance. In December 2024, the UK’s biometrics and surveillance camera commissioner, William Webster, raised alarms over police retaining and using images of individuals arrested but never formally charged. Recently, South Wales Police settled a claim by a black man, wrongfully arrested and held for over 13 hours due to the technology.

Iain Gould, Choudhury’s attorney at DPP Law, emphasized that law enforcement must ensure artificial intelligence does not replace human judgment. Instead, it should be utilized only alongside careful human oversight.

The Home Office indicated that it is reviewing guidance and training aimed at minimizing errors and ensuring public trust in the use of retrospective facial recognition. A new national facial matching system is also in development, featuring an improved algorithm set to undergo independent testing.