“Safe City” is a vendor term for integrated urban surveillance: networks of cameras, facial and license-plate recognition, and command centers sold to governments as crime-fighting tools. Huawei is the best-known exporter, with dozens of agreements across Asia, Africa, Latin America and Europe, but US, Japanese and Israeli firms also supply surveillance technology worldwide. Independent research has found little verified evidence that these systems deliver the large crime reductions vendors advertise, while documented misuse against opposition figures and activists is a recurring concern.
What a “Safe City” system includes
There is no single standard. According to CSIS, Huawei markets facial and license-plate recognition, social media monitoring and other surveillance capabilities under the “Safe City” label (CSIS, 2019). Privacy International found that what is delivered varies widely from city to city, ranging from CCTV hardware and AI analytics to fiber-optic networks (Privacy International, 2021).
Typical components include:
- Large camera networks feeding a central police command center
- Facial recognition and automatic number plate recognition (ANPR)
- Data platforms that link video with other police or government databases
- Financing packages, often from Chinese state lenders, bundled with the hardware
How far the technology has spread
The global picture
The most-cited dataset remains Carnegie’s AI Global Surveillance Index from 2019. Author Steven Feldstein found that at least 75 of 176 countries were using AI surveillance, with 56 deploying smart or safe city platforms and 64 using facial recognition (Carnegie, 2019).
| Supplier | Countries supplied (2019) | Source |
|---|---|---|
| All Chinese companies | 63 | Carnegie, 2019 |
| Huawei | at least 50 | Carnegie, 2019 |
| All US companies | 32 | Carnegie, 2019 |
| NEC (Japan) | 14 | Carnegie, 2019 |
| IBM (US) | 11 | Carnegie, 2019 |
| Palantir (US) | 9 | Carnegie, 2019 |
| Cisco (US) | 6 | Carnegie, 2019 |
Adoption is not limited to autocracies: Carnegie found 51% of liberal democracies used AI surveillance. Western companies are also significant suppliers, even if no single one matches Huawei’s reach.
Huawei itself has claimed more than 160 smart city projects in over 100 countries and regions, a company figure that Privacy International reported but could not independently verify (Privacy International, 2021).
Who buys Chinese systems
CSIS counted 73 Huawei Safe City agreements in 52 countries. Of these, 71% were in countries rated “partly free” (44%) or “not free” (27%) by Freedom House, and 59% were in Asia or sub-Saharan Africa (CSIS, 2019).
Economists Martin Beraja, Andrew Kao, David Yang and Noam Yuchtman tested this pattern with global trade data on facial recognition AI. They found that autocracies and weak democracies were more likely to import surveillance AI from China, especially in years of domestic unrest, and that no such political bias appeared in AI imports from the US (Brookings, 2023).
Documented deployments
| Country | Vendor / project | Documented details | Source |
|---|---|---|---|
| Nigeria | Chinese suppliers | Over $470 million on facial and number-plate recognition; largest spender among 11 African countries studied | IDS, 2026 |
| Mauritius | Chinese suppliers | About $456 million | IDS, 2026 |
| Kenya | Huawei (Nairobi Safe City) | About $219 million; Nairobi system of 1,800 cameras | IDS, 2026; Africa Center, 2020 |
| Uganda | Huawei | $126 million police CCTV with facial recognition; ~2,500 of 3,200 cameras installed in Kampala | Reuters, 2019 |
| Serbia | Huawei (Belgrade); Dahua (municipalities) | About 8,000 Huawei cameras with facial recognition capability planned for Belgrade; at least 10 municipalities bought Dahua cameras with facial recognition | RFE/RL, 2023 |
| Ecuador | Chinese-built ECU-911 | 4,300 cameras and 16 monitoring centers; a feed reached the national intelligence agency | The New York Times, 2019 |
| France, Spain | Huawei smart city projects | Valenciennes; Rivas Vaciamadrid and Barcelona; some cities did not answer transparency requests | Privacy International, 2021 |
Evidence on crime reduction
Vendor claims vs. independent data
The clearest test case is Nairobi. Huawei claimed crime in areas covered by its system fell 46% in 2015 compared with the previous year. CSIS compared that claim with Kenyan National Police Service statistics, which showed smaller decreases in Nairobi, slight increases in Mombasa, and Nairobi crime back above pre-installation levels by 2017 (CSIS, 2019). CSIS described some Huawei marketing figures as “grossly exaggerated.”
The 2026 IDS study of 11 African countries reached a similar conclusion, reporting no compelling evidence that smart surveillance had reduced terrorism or serious crime (IDS, 2026).
What broader CCTV research shows
A 40-year systematic review and meta-analysis by Piza, Welsh, Farrington and Thomas found CCTV associated with a significant but modest drop in crime, with the strongest effects in car parks and residential areas and against vehicle and property crime. Actively monitored systems and those combined with other measures performed better (Piza et al., 2019). Those modest, context-specific effects fall well short of the sweeping citywide results promised in sales material.
Civil liberties case studies
Uganda
In August 2019 The Wall Street Journal reported that Huawei technicians had helped Ugandan and Zambian officials intercept encrypted communications and track political opponents, including Ugandan opposition figure Bobi Wine. The Journal found no evidence that Huawei executives in China knew of or approved the activities, and Huawei denied wrongdoing (CNBC, 2019). The same month, opposition politicians told Reuters they feared the new Kampala camera network would be used against protesters ahead of the 2021 election (Reuters, 2019).
Serbia
Serbia shows how deployments can run ahead of legislation. RFE/RL reported in 2023 that Belgrade’s Huawei facial recognition capability had not been activated because no legal framework allowed it, and that the government had withdrawn draft police laws authorizing the technology after public pressure, most recently in December 2022 (RFE/RL, 2023). EDRi documented an earlier 2021 draft that would have enabled mass biometric surveillance (EDRi, 2021).
The Serbian cases also involve non-Chinese technology. In December 2024, Amnesty International reported that Serbian police and intelligence used forensic tools from Israeli company Cellebrite to unlock activists’ and journalists’ phones before installing spyware (Amnesty International, 2024). In February 2025, Cellebrite said it would stop certain Serbian customers from using its products (JURIST, 2025).
Ecuador
A 2019 New York Times investigation found that Ecuador’s Chinese-built ECU-911 system, sold as an emergency and crime-response network, also sent footage to the country’s domestic intelligence agency (The New York Times, 2019).
What this means for you
- In cities with Safe City projects, assume public cameras may be linked to recognition systems, even where the legal basis is unclear.
- Travelers and journalists visiting countries with documented misuse should plan for both physical and digital surveillance: minimize data carried on phones, use full-disk encryption and strong passcodes, and be aware that devices taken by authorities may be forensically examined.
- A VPN does not protect you from cameras. It encrypts internet traffic, but face and license-plate recognition work on what is visible in public.
- Check what your city has bought. In Europe, freedom-of-information requests and data protection authorities can reveal it, and civil society groups such as SHARE Foundation in Serbia and Privacy International use them regularly.
- In the cases above, deployment came before regulation. Clear laws on retention, oversight and facial recognition are the main protection citizens have, so they are worth supporting.
Methodology
Sources are think-tank datasets (Carnegie, CSIS), academic research (Beraja et al. via Brookings; Piza et al.), a 2026 multi-country study by the Institute of Development Studies and the African Digital Rights Network, and investigative reporting by Reuters, The Wall Street Journal (as reported by CNBC), The New York Times and RFE/RL. Human rights findings come from Amnesty International, Privacy International and EDRi.
Safe City contracts are often confidential, so all counts are minimums based on public records and reporting. The Carnegie and CSIS datasets date from 2019 and have not been comprehensively updated; current deployments are likely higher. Spending figures for African countries come from the IDS study and may combine several contracts. Company figures, such as Huawei’s project count, could not be independently verified.