Originally published 2020-10-23.

Pictured left: Portland activist Christopher Howell, who began building a facial recognition tool after being tear-gassed at a protest. Photograph by Mason Trinca for The New York Times.
Facial recognition is usually discussed as a tool used by governments and police to identify members of the public. In 2020, during the wave of protests that followed the murder of George Floyd, some activists turned it around. Frustrated that officers at demonstrations were covering their names or badge numbers, they began experimenting with software to identify the police themselves.
The best-known example, reported by The New York Times in October 2020, was Christopher Howell, a self-taught programmer in Portland, Oregon. Similar projects had already appeared in Hong Kong and France. Here is what happened, how the technology works, and why the idea raises difficult questions on both sides.
Christopher Howell's project in Portland
Portland saw months of nightly protests in 2020, and clashes with police and federal agents were frequent. Howell, who had been tear-gassed at a demonstration, said he noticed that many officers were not displaying identifying information. Drawing on his knowledge of neural networks, the type of machine-learning system behind most modern facial recognition, he began building a database of officers' faces gathered from publicly available photos.
His reasoning was about accountability. If an officer used excessive force while unidentifiable, he argued, complaints and lawsuits would go nowhere. "Accountability is important. We need to know who is doing what, so we can deal with it," he told The New York Times.
At the time of the reporting, the tool was a work in progress rather than a finished product, and it was unclear how accurate it was or how widely it would be used.
Earlier efforts in Hong Kong and France
Howell was not the first to try this. During the 2019 Hong Kong protests, when riot police often operated without visible identification, a local activist named Colin Cheung developed software aimed at identifying officers from photographs. The same protests saw demonstrators tear down "smart" lampposts over fears that they carried cameras feeding government surveillance.
In France, artist Paolo Cirio went further. In October 2020 he published Capture, a project showing about 4,000 faces of French police officers taken from protest photographs and processed with facial recognition software. He invited the public to help identify them. Contrary to some early reports, Cirio's stated aim was not to expose wrongdoers but to provoke debate: the project was designed to promote his campaign for a Europe-wide ban on facial recognition, by showing the police what it feels like to be targeted. France's interior minister, Gérald Darmanin, and police unions pushed for it to be taken down, and its removal from an exhibition at the Le Fresnoy arts centre was widely described as censorship.
Why facial recognition is so controversial
Facial recognition systems compare a face captured in an image with a database of known faces and suggest likely matches. They have become cheap and powerful, and they are used for everything from unlocking phones to scanning crowds. Critics point to several problems.
- Errors and bias. No system is 100% accurate. The Gender Shades study by Joy Buolamwini and Timnit Gebru found that some commercial systems misclassified the gender of darker-skinned women up to 36% of the time, compared with under 2% for lighter-skinned men. Accuracy has improved since, but results still depend heavily on image quality and on the data used to train the software.
- Wrongful arrests. In the United States, several people have been arrested after faulty matches, including Robert Williams in Detroit in 2020, who was detained for a theft he did not commit.
- Chilling effect on protest. In China, facial recognition is part of a vast surveillance system, and human rights groups have criticised its use against Uyghurs and dissidents. Knowing that cameras can log every face can deter people from attending demonstrations.
- Uneven use. Campaigners argue that cameras and searches are often concentrated in poorer and minority neighbourhoods, which can amplify existing inequalities in policing.
Concerns like these led San Francisco to become the first major US city to ban facial recognition use by police and city agencies in May 2019. Portland followed with its own ban on municipal use in September 2020.
Covered badges and the accountability gap
The activists' projects grew out of a real problem. Officers who cannot be identified are much harder to hold to account, whether through internal complaints, courts or the press. During the 2020 protests, there were widespread reports of officers in several US cities covering name tags or badge numbers, and federal agents in Portland wore uniforms without individual names. Some police departments said this protected officers from harassment online, while critics said it undermined public trust.
Many police forces require officers to display a name or number while on duty, and the question of how to enforce such rules during protests has been debated in several US cities since 2020. Identification requirements are a simpler and more reliable route to accountability than crowdsourced face matching.
The risks of turning the tool around
Using facial recognition against police does not make the technology's flaws disappear. A possible match is only that: a suggestion. Lookalikes, blurry photos, masks and helmets all increase the chance of error. If an activist tool wrongly names an officer, an innocent person could be accused of violence in public, with real consequences for them and their family.
There are also safety concerns. Some activists in 2020 went beyond identification and shared officers' home addresses, family details or political views online. Police groups warned that this kind of exposure, often called doxxing, could lead to threats or attacks. Publishing private information can also break the law in some places.
Supporters reply that the public faces exactly these risks every day from police and commercial facial recognition, often with fewer safeguards. That, in a sense, was Paolo Cirio's point: if the technology feels invasive when it is aimed at officers, it is just as invasive when it is aimed at everyone else.
What this tells us about surveillance technology
The episode shows how quickly powerful tools spread once they become cheap and widely available. Facial recognition is no longer the preserve of governments or large companies; anyone with programming skills and a collection of photos can attempt it.
That raises a broader question for lawmakers. Rules that focus only on police use leave gaps when private citizens, companies or activists deploy the same technology. Many experts argue for clear limits that apply to everyone, independent testing of accuracy, and a firm principle that a software match should never be treated as proof without other evidence.
For readers, the practical lesson is simple: treat any claimed identification, whether made by police software or by an activist's app, with caution until it is backed up by independent evidence.
Sources: The New York Times (21 October 2020): activists turn facial recognition on police; Wikipedia: Paolo Cirio; Wikipedia: Facial recognition system; Wikipedia: Murder of George Floyd.

COMMENTS