Daily life in China is gated by security technology, from the body scanners and X-ray machines at every urban metro station to the demand for ID numbers on social media platforms so that dangerous speech can be traced and punished. Technologies once seen as potentially empowering the public have become tools for an increasingly dictatorial government—tools that Beijing is now determined to sell to the developing world.

In 2015, the Chinese government launched its Made in China 2025 plan to dominate cutting-edge technological industries. This was followed up last year for plans for the country to be a world leader in the field of artificial intelligence by 2030 and to build a $150 billion industry. The developing world is a big part of these ambitions. But China doesn’t just want to dominate these markets. It wants to use developing countries as a laboratory to improve its own surveillance technologies.

Zimbabwean President Robert Mugabe, left, and his Chinese counterpart Xi Jinping participate in a signing ceremony at the Great Hall of the People on Aug. 25, 2014, in Beijing, during Mugabe’s five-day state visit. (Diego Azubel-Pool/Getty Images)

Many parts of Africa are now essentially reliant on Chinese companies for their telecoms and digital services. Transsion Holdings, a Shenzhen-based company, was the No. 1 smartphone company in Africa in 2017. ZTE, a Chinese telecoms giant, provides the infrastructure for the Ethiopian government to monitor its citizens’ communications. Hikvision, the world’s leading surveillance camera manufacturer, has just opened an office in Johannesburg.

The latest is CloudWalk Technology, a Guangzhou-based start-up that has signed a dealwith the Zimbabwean government to provide a mass facial recognition program. The agreement is currently on hold until Zimbabwe’s elections on July 30. But if it goes through, it will enable Zimbabwe, a country with a bleak record on human rights, to replicate parts of the surveillance infrastructure that have made freedoms so limited in China. And by gaining access to a population with a racial mix far different from China’s, CloudWalk will be better able to train racial biases out of its facial recognition systems—a problem that has beleaguered facial recognition companies around the world and which could give China a vital edge.

The CloudWalk deal is built on the back of a long-standing relationship between former Zimbabwean President Robert Mugabe’s regime, seen by China as an ideological ally, and Beijing. Current President Emmerson Mnangagwa was sworn into office in November 2017 after a military coup forced Mugabe to resign after 37 years of increasingly repressive rule. But activists fear that Mnangagwa, Mugabe’s former consigliere, will continue the patterns of his predecessor, especially if his regime is backed up with new security technology.

The deal between CloudWalk and the Zimbabwean government will not cover just CCTV cameras. According to a report in the Chinese state newspaper Science and Technology Daily, smart financial systems, airport, railway, and bus station security, and a national facial database will all be part of the project. The deal—along with dozens of other cooperation agreements between Harare and Chinese technology and biotech firms—was signed in April. Like every other foreign deal done by a Chinese firm of late, it has been wrapped into China’s increasingly all-encompassing Belt and Road Initiative.

Like every other foreign deal done by a Chinese firm of late, it has been wrapped into China’s increasingly all-encompassing Belt and Road Initiative.

The CloudWalk deal is the first Chinese AI project in Africa. Google is opening its first Africa AI research center in Ghana this year, but Eric Olander, founder of the China Africa Project—a podcast and online resource that examines the relationship between China and Africa—noted that many Western companies “aren’t willing to make that step that the Chinese are willing to do. … [The Chinese] are willing to make an investment in a market as volatile as Zimbabwe, where companies from other countries are not.”

Indeed, with massive state and private backing for AI projects—according to a CB Insights report, nearly half of global investment in AI went to Chinese start-ups last year, surpassing the United States for the first time—Chinese companies can afford to take risks. CloudWalk itself was the recipient of a $301 million grant from the Guangzhou municipal government.

“We are concerned about the deal, given how CloudWalk provides facial recognition technologies to the Chinese police,” said Maya Wang, a senior China researcher for Human Rights Watch. “We have previously documented [the Chinese] Ministry of Public Security’s use of AI-enabled technologies for mass surveillance that targets particular social groups, such as ethnic minorities and those who pose political threats to the government.”

Some Zimbabweans are concerned about how their data will fare in China. Andy, who asked that only his first name be used, is studying for a Ph.D. at Beijing Normal University. For him, “the question is what the Chinese company will do with our identities. … It sounds like a spy game.” He also says that he “know[s] for a fact” that “the Zimbabwe government will use this tech to try and control people’s freedom.”

In Zimbabwe, freedom of expression has long been curtailed or monitored by various means. In 2015, Mugabe accepted a gift of cybersurveillance software from the Iranian government, including IMSI catchers, which are used to eavesdrop on telephone conversations. In 2016, he cited China as an example of social media regulation that he hoped Zimbabwe could emulate.

Communications are routinely monitored and censored in China; poorly chosen words sent on WeChat, the main messaging platform, can land the sender in jail. This principle of intrusive surveillance is well established in Zimbabwe—last year’s Cybercrime and Cybersecurity Bill, for example, criminalized communicating falsehoods online, legal rhetoric that has been used in China to silence dissent. With the technology provided by the CloudWalk deal, government opponents will have even fewer places to hide.

Zimbabwe is not the only African country that seeks to emulate China’s model. Tanzania, a country that routinely harasses government critics, passed a law last year regarding online content that seems heavily influenced by the Chinese model. The Tanzanian law prohibits the posting of “false content,” just as “making falsehoods” is banned under Chinese law. The vague idea of “content that causes annoyance” is now verboten in Tanzania, just as “destroying the order of society” is in China. The Tanzanian government justified this law as a means of cracking down on “moral decadence”; China has previously banned “decadent” material from social media.

Wang notes that AI-powered recognition systems are not “fundamentally abusive.” CloudWalk agrees. “Whether or not privacy is ultimately violated depends on how the Zimbabwean government uses it,” said Fu Xiaolong, a spokesman for the company, adding that CloudWalk would not be responsible for any such decisions. “Just as the United States sells arms around the world—it does not care whether other governments use American weapons to kill people.”

The purported use of this kind of surveillance is to reduce harm, not increase it. Whether or not the technology is activated, the panopticon effect of visible surveillance—especially when labeled as facial recognition—has been claimed to reduce crime.

Whether or not the technology is activated, the panopticon effect of visible surveillance—especially when labeled as facial recognition—has been claimed to reduce crime.

Olander stresses that there is a “real need” for “genuine security measures” in Zimbabwe. “The new administration has been far more open than Robert Mugabe’s, so while I am concerned that this technology will be used for nefarious means … we should also give the benefit of the doubt that maybe the security services will use it for legitimate uses,” he said. But activists on the ground in Zimbabwe aren’t so sure. “Past experience has also shown that the Zimbabwean government will use a legitimate excuse, for example, to preserve law and order as a guise to actually monitor and restrict citizens’ online expression and behaviors,” said Kuda Hove, a program officer for the Media Institute of Southern Africa, which promotes freedom of expression in Zimbabwe.China is not the only country to sell surveillance technology to potentially abusive governments. A report by Security for Sale, an international consortium of journalists, found that between 2014 and 2017, almost one-third of licenses granted by the European Union to export surveillance products were to countries deemed “not free” by the watchdog Freedom House. The Zimbabwe deal is not unique—in 2016, Beijing donated $14 million worth of similar equipment to Ecuador, for example, which the Ecuadorian government claims has helped cut crime rates by 24 percent.

Countries that have received Chinese technology are not able necessarily to mimic the extent of Chinese mass surveillance. “The entire surveillance infrastructure is no small feat: the systems and multilayered and numerous. To achieve the kind of surveillance state the Chinese government has, one also needs to implement, for example, a national identification card system and a real name registration system,” Wang noted. It’s also backed up, as Paul Mozur recently examined in the New York Times, by enormous amounts of manpower; the vast majority of surveillance in China is carried out by human watchers, not AI systems.

The Zimbabwe deal is unique in that as part of the agreement—the value of which CloudWalk declined to share—Harare will send data on millions of black faces to the Chinese company to help train the technology toward darker skin tones. The currency here is data as well as dollars.

“People did not consent to the use of their biometric data in this way,” Hove said. “Unfortunately, people do not have any way of holding the government accountable as there are no laws in place or any regulatory body tasked with the protection of people’s privacy or data protection.” Zimbabwe’s 2002 Access to Information and Protection of Privacy Act doesn’t cover biometric data or cross-border flows of data, and, as Hove notes, “the government has rarely ever acted in the people’s interests.”

The training of AI to work better on black faces is a significant opportunity for CloudWalk and for Chinese AI in general. A recent study by the Massachusetts Institute of Technology’s Media Lab tested the accuracy of three major facial recognition software providers: Microsoft, IBM, and Megvii (a Chinese company). The study found that when attempting to identify the gender of a person from a picture, the error rate for lighter-skinned men was less than 1 percent, while for darker-skinned women it was up to 35 percent. This is because the accuracy of AI depends on the data it is fed, and, worldwide, facial recognition AI has been trained on predominantly white and male faces, even inside China.

There have been other racial upsets in the industry over the years—in 2015, Google apologized after its Google Photos app mistakenly labeled black people as gorillas. Brian Brackeen, the black founder of the Miami-based facial recognition company Kairos, recently said coded bias in algorithms meant that such technology should never be used by law enforcement. Facial recognition has largely failed in practice in open environments in the West; trials in the United Kingdom by police have resulted in 91 percent of cases being false positives.

Facial recognition has largely failed in practice in open environments in the West; trials in the United Kingdom by police have resulted in 91 percent of cases being false positives.

Quite how successful the system is in China is unclear; there are no independent studies or government watchdogs. But the government is reportedly building the world’s largest facial recognition database, using the access it has to the facial data of its 1.3 billion citizens. There is already some racial diversity within this, especially thanks to the mass surveillance and incarceration programthat the government has implemented in the western province of Xinjiang, where the population is predominantly Muslim Uighurs of Turkic descent.

But with the CloudWalk deal, a Chinese company now has access to a data pool far more diverse than China’s ethnic makeup. So, while the government in Harare has paid out for the technology from CloudWalk, it’s likely that the company will benefit in the long term from this opportunity to rapidly improve its systems. Zimbabwean leaders are “getting access to technology and tools that they would never be able to afford on the open market if they didn’t have a currency other than the data of their own people to leverage against that,” Olander said. “For Natasha Msonza, the co-founder of the Digital Society of Zimbabwe, “it feels like [CloudWalk] is looking for guinea pigs,” she said, adding, “I don’t believe that the Zimbabwe government gave this proposition much thought before volunteering its citizens to be subjected to racial facial recognition experiments.”

Other than increased security and surveillance measures, Zimbabweans will not see any return on the research that their personal data has helped accelerate. Acceleration is the whole point because the global AI race is ultimately a race to set standards. The Chinese government defines its ambitions as becoming the country that is “setting the pace.” As racial upsets in facial recognition have shown, the standards in this field are still to play for. But with unprecedented access to a more diverse range of data, Chinese companies are edging ever closer to this goal—spreading their model of authoritarianism along the way.

Amy Hawkins is a freelance writer based in Beijing. @XLHawkins