From The Big Story: How a Research Blog Took on Big Surveillance in China—and Won

At a location he keeps secret, John Honovich was on his laptop, methodically scouring every link on a website for a conference half a world away.

Hikvision, the world's largest security camera manufacturer, was hosting the event, the two thousand eighteen AI Cloud World Summit, in its hometown of Hangzhou, a city of about ten million people not far from Shanghai.

Honovich, the founder of a small trade publication that covered video surveillance technology, wanted to find out what the latest Hikvision gear could do.

He zeroed in on one section of the conference agenda titled Eco-Friendly, Peaceful, Relaxed, and found a description of an AI-powered system installed around Mount Tai, a historically sacred mountain in Shandong.

A video showed Hikvision cameras pointed at tourists climbing the thousands of stone steps leading to the famous peak.

Piano music played as a narrator explained, in Mandarin with English subtitles, that the cameras were there to identify all visitors to ensure the safety of all.

The video cut to a shot of a computer screen, and Honovich hit pause.

He saw a zoomed-in view of one visitor's face.

Below it was data that the camera's AI had inferred.

Honovich downloaded the video and took screenshots of the computer screen for safekeeping.

Later, with the help of a translator, he scrutinized every bit of text on that screen.

One set of characters, the translator explained, suggested each visitor was automatically sorted into categories.

Age, sex, wearing glasses, smiling.

When Honovich pointed at the fifth category and asked, "What's this?" the translator replied, "Minority." Honovich pressed, "Are you sure?" The translator confirmed there was no other way to read it.

Honovich was shocked.

In his many years in the industry, he'd never seen a surveillance company set out to automatically detect racial minorities.

The feature seemed completely unethical to him, and he immediately wondered how China might use it against the Uyghur people, a mostly Muslim ethnic minority group in the province of Xinjiang.

Honovich had seen reports trickling out in the West of Uyghurs being selected to constrictive surveillance and mass detentions.

Clicking through the AI Summit website, Honovich couldn't tell whether Chinese authorities were using this technology to oppress minorities, but he saw that danger coalescing.

He quickly wrote up an article about Hikvision's ethnicity detection technology, including the video, screenshots, and a no comment from the company, and posted it on the website of IPVM, the trade publication he had founded.

He talked about the discovery with one of IPVM's reporters, Charles Rollet, a Frenchman who lives outside the U.S. and also keeps his location secret.

Rollet had written about how Hikvision and Dahua, the second largest video surveillance manufacturer in China, were reaping huge profits from government work in Xinjiang.

Rollet had a newspaper background, and though he was twenty-five, talked like an ink-stained newsie twice his age, all scoops and calling out abuses and hard-hitting news.

By trawling through publicly available materials online, Rollet had learned that Hikvision had landed a deal to build a mass face recognition system to cover one Xinjiang county, including a re-education center and some of its mosques, and a contract to install video conferencing systems in mosques, presumably so attendees could watch sermons broadcast by the government.

Dahua won the bigger contract, six hundred and eighty-six million dollars to build camera-equipped police stations in another part of Xinjiang.