SpyShop Europe Research · 2026 Investigation

2026 City Surveillance Index

Often expanded and rarely audited, we quantify the real-world trade-off between surveillance camera number growth, crime trends, and public trust across 100 cities.

About this study

The Cost of the Surveillance Bargain in 2026

At SpyShop Europe we have been close to the surveillance and counter-surveillance industries for nearly two decades. In that time we have watched governments install hundreds of millions of cameras across European capitals, transit hubs, and public squares. The build-out was sold to citizens as a fair trade: some loss of privacy in exchange for a measurable gain in safety. For years the bargain was widely accepted, and the cameras kept coming.

But did the promised safety actually arrive? And has public trust in how the state handles its citizens' data kept pace with the speed at which the cameras have been collecting it? Those are the two questions this dataset sets out to answer, jurisdiction by jurisdiction.

To carry out the study we tracked three indicators over the same reference window. For cameras we compared industry market estimates of national CCTV stock in 2015 against the most recent year available, sourced from Comparitech 2025, IHS Markit / Omdia, BSIA for the United Kingdom, and national security-industry associations elsewhere. For crime we used the most recent year of total police-recorded offences against the 2015 figure, drawn from national statistical offices or, for the European Union, from Eurostat. For trust we used the relative change in the proportion of citizens reporting that they are not concerned about how the state handles their personal data, drawn from Pew Research for the United States and Anglosphere, Special Eurobarometer 487a then 551 across the EU, and the Edelman Trust Barometer for the rest of the surveyed world.

Police-recording quality varies across countries, and where the change in recording practice itself accounts for a meaningful share of the apparent crime trend (as it does for England and Wales, where HMICFRS assessed compliance rising from 80.5% to 94.8% across the window) we flag the caveat directly. For jurisdictions whose crime and trust series are produced by state-controlled bodies without independent statistical oversight (Russia, China, much of the Gulf, and parts of Sub-Saharan Africa), we publish the camera figures but mark the crime and trust cells as not independently verifiable. Below you can compare any two cities side by side or sort the full 100-city dataset by any column.

Instructions for journalists

How to use this dataset.

A short guide to what the data supports, what it does not, and how to attribute every figure in this report.

“The tables below document government-operated CCTV camera density in 100 cities and direction-of-travel indicators for recorded crime and public trust over the reference window (2015 to most recent available, typically 2024 or 2024/25). All figures cited inline are the values published by the original source agency. Direction codes and scope badges follow the taxonomy defined in § 3.5 of the methodology.”

Compare two cities

Pick any two cities and view their indicators side by side.

London versus Tokyo. Stockholm versus Singapore. Berlin versus Beijing.
The dataset reads differently when two jurisdictions sit next to each other.

All 100 cities are available in the pickers. The panel uses the same diverging heatmap as the full table below, so the visual contrast between any two rows reads identically across the two views.

vs
London
United Kingdom · Europe
Washington D.C.
United States · Americas
The dataset

The 100-city dataset, with Since-2015 deltas.

Ten columns. All sortable. Heat-mapped.
Every value traces to a named public source, every direction-of-travel is calculated from endpoint figures you can verify, and every cell is built to be quoted by row.

Click any header to sort. All three Since-2015 deltas are % change vs 2015: cameras from industry market estimates (Comparitech, IHS Markit, BSIA, national associations); crime from national statistical offices or Eurostat; trust from Pew (US, Anglosphere), Special Eurobarometer 487a then 551 (EU), Edelman Trust Barometer (rest of world). Background colour encodes both direction and magnitude on a diverging scale, comparable across all three columns.

Since 2015
# City Country Region Cameras / 1K People Cameras / km² Cams Crime Trust Privacy Law

How the Since-2015 deltas are computed.

Cameras.
National stock estimate 2025 divided by national stock estimate 2015, less 1. Sourced from industry market reports (IHS Markit / Omdia, Comparitech 2025, BSIA for the UK, national security-industry associations elsewhere). Override with city-level estimate where one exists: London, Stockholm, Paris, Moscow, the four Chinese megacities, Dubai, Riyadh, Hyderabad, Seoul.

Crime.
Most recent year of total police-recorded offences divided by the 2015 figure, less 1. Sourced from national statistical office or Eurostat crim_off_cat. City overrides where the city-series is firm: London (MOPAC), Berlin (PKS), Stockholm (BRÅ), Paris (SSMSI), Tokyo (NPA), Singapore (SPF), Washington DC (MPD), New York (NYPD), LA, Chicago.

Trust.
Relative change in the proportion of citizens not reporting concern about state data handling. Pew Research for US / UK / Canada / Australia / NZ. Special Eurobarometer 487a (2019) vs 551 (2024) / 553 (2025) for EU. Edelman Trust Barometer for the rest of the surveyed world.

Cells showing n/a are jurisdictions with no comparable time series, overwhelmingly state-controlled crime statistics (Russia, China, much of the Gulf and Sub-Saharan Africa) where the published figures cannot be independently validated.

Confidence.
Country-level estimates carry ±20% uncertainty; city overrides are tighter (typically ±5 to 10%). Replace the camsD / crimeD / trustD fields in the cities array with project-dataset values to tighten any row.

Reading the cells.
Background colour encodes both direction and magnitude on a diverging scale: amber for positive change (deepening at +50%, +150%, +300%), red for negative (deepening at -15%, -30%), neutral for n/a. The same scale applies across all three Since-2015 columns so cells stay directly comparable.

When the camera is wrong

The build-out is institutional. The cost is personal.

Documented cases of facial recognition misidentification, biometric overreach, and GDPR enforcement actions, drawn from court records and named press reporting.

When systems trained on imperfect data make decisions about real people, the errors do not distribute evenly. The cases below are the ones that reached the public record.
Misidentification · United States
Robert Williams
Detroit, January 2020
Wrongfully arrested in front of his family and detained for thirty hours after a facial recognition system matched a low-quality CCTV still to his driving licence photo. The first publicly documented wrongful arrest in the United States caused by facial recognition error. The case was settled by the City of Detroit in 2024 with a $300,000 payout and binding restrictions on how Detroit Police can use the technology in future investigations.
Source: ACLU v City of Detroit settlement, 2024
Misidentification · United States
Porcha Woodruff
Detroit, February 2023
Eight months pregnant when she was arrested in front of her two daughters and held for eleven hours, accused of a carjacking. She had been identified by the same facial recognition system that misidentified Williams three years earlier. The case prompted a US Federal Trade Commission inquiry into how police forces validate facial recognition matches before making arrests.
Source: New York Times, August 2023; FTC inquiry filing 2024
GDPR enforcement · Sweden
Skellefteå municipality
€20,000 fine, August 2019
A school in Skellefteå piloted facial recognition to track student attendance. The Swedish data protection authority ruled that the consent obtained from parents was invalid because the imbalance of power between school and pupil meant consent could not be freely given. The first GDPR fine in Sweden, and a precedent now widely cited across European education and workplace cases.
Source: Datainspektionen ruling, 21 August 2019
GDPR enforcement · Italy
Clearview AI
€20,000,000 fine, March 2022
The Italian data protection authority fined Clearview AI €20 million for unlawfully scraping and indexing photographs of people in Italy from public websites and social media to populate a facial recognition database sold to law enforcement and private clients. Clearview was ordered to delete all Italian-resident data and prohibited from further processing. Comparable rulings followed from authorities in France, the United Kingdom and Greece.
Source: Garante per la protezione dei dati personali, 9 March 2022
Biometric overreach · United Kingdom
Ed Bridges
Cardiff, Court of Appeal 2020
Bridges, a civil liberties campaigner, brought the world's first successful legal challenge to police use of live facial recognition. South Wales Police had scanned his face twice, once in a city centre and once at a peaceful protest. The Court of Appeal ruled the deployment unlawful on three counts, including a failure of due diligence on bias in the system. UK police forces revised their policies; the deployment continues elsewhere.
Source: Bridges v South Wales Police, EWCA Civ 1058 (2020)
Workplace surveillance · Spain
López Ribalda v Spain
European Court of Human Rights, 2019
A supermarket in Catalonia installed hidden cameras to identify employees suspected of theft. Five cashiers were dismissed on the basis of the recordings. The European Court of Human Rights ruled in 2019 that, in the circumstances, the covert surveillance did not violate Article 8 of the European Convention on Human Rights. The judgment is now the leading authority on employer covert workplace surveillance across the Council of Europe's 46 member states, and is contested in current EU AI Act case law.
Source: ECtHR Grand Chamber, López Ribalda and Others v Spain, October 2019
What These Cases Have In Common
Each began as a deployment of technology that, on its institutional face, had a legitimate purpose. The harm appeared at the point where the system met an actual human being whom it categorised, identified, or recorded. The legal framework caught up only after the harm was made public. There is no central register of cases where it didn't.
China exception

China: A Note on What the Index Counts

Chinese cities appear mid-table in the 100-city ranking. Whether that accurately represents the scale of state camera access in China depends on how "government-accessible" is defined.

Total CCTV cameras, China
~700 million
Estimates range from 540M (IHS Markit 2021 projection) to 700M+ (Comparitech, multiple government sources). China holds over half of all surveillance cameras on Earth.

Sources: IHS Markit, Comparitech, Associated Press

Camera-to-citizen ratio
1 : 2
Approximately one camera for every two people. The national average works out to ~494 cameras per 1,000 people, roughly 37× the density of London.
Travel restrictions (2018)
23 million
Attempts to purchase flight (17.5M) and high-speed rail tickets (5.5M) blocked in 2018 via court-ordered judgment-defaulter blacklists, per China's National Public Credit Information Center.

Source: AP / National Public Credit Information Center 2019 report

Facial recognition capability
200M+ AI cameras
At least 200 million cameras are equipped with AI-powered facial recognition, capable of identifying individuals in seconds across city-wide networks.

Social Credit System: Structure

Popular accounts of China's Social Credit System often describe a single national score in which every citizen starts with 1,000 points, earns rewards for good behaviour, and loses them for bad, with "D ratings" and "AAA ratings" tied to real-world privileges. Research institutions including MERICS, MIT Technology Review, and Stanford's Freeman Spogli Institute describe the operating reality differently: no unified national numerical score exists.

The system as it currently operates has three distinct components: corporate compliance ratings (the largest and most developed component, applied to businesses), court-ordered judgment-defaulter blacklists (nationally enforced, producing the widely-reported travel bans), and regional pilot programmes in roughly 40 cities. The 1,000-point, A-to-D grading model originates in one of those pilots, Rongcheng, a city of 740,000 in Shandong Province, and has not been scaled to provincial or national level.

The nationally-enforced judgment-defaulter blacklist produces the most measurable outcomes: in 2018, roughly 23 million ticket purchase attempts were blocked per China's National Public Credit Information Center. Sector-specific blacklists restrict access to jobs, loans, and government services. The underlying infrastructure of cameras, AI analytics, and mandatory real-name payment systems operates independently of whether a single numerical score exists.

Penalties Consequences of being blacklisted
23M
ticket purchase attempts blocked in 2018 alone, the most measurable real-world effect of the system to date.
  • Travel restrictions Blocked from flight and high-speed rail tickets (court-ordered)
    National
  • Employment limits Blocked from government jobs and state-owned enterprise roles
    National
  • Housing and credit Restricted access to mortgages and certain financial products
    Sector
  • Public naming Identities published on official blacklist websites
    National
  • Children's schooling In some pilot areas, barred from private schools
    Pilot cities
  • Slower services Reduced internet speeds and slower processing at government offices
    Pilot cities
Rewards Perks in pilot programmes
~40
cities run pilot reward schemes of various kinds, tied to local regulation, not a unified national score.
  • Utility discounts Reduced bills for high-scorers in Rongcheng and similar pilots
    Pilot cities
  • Credit terms Preferential rates from banks in participating schemes
    Sector
  • Fast-track services Expedited processing at government offices
    Pilot cities
  • Public recognition "Model citizen" status published locally
    Pilot cities
  • Deposit waivers Bike-share, library, and rental programme benefits
    Pilot cities

Sources: Stanford Freeman Spogli Institute (Meritown SCS study, 2024), MERICS, MIT Technology Review, Brussee (2023). Scope tags indicate whether a consequence applies under national law, by sector (banking, courts, employers), or only in the ~40 pilot-city programmes.

The Surveillance Infrastructure

China's camera network operates through three integrated programmes:

Programme 01 · Urban
天网SkyNet, City-wide monitoring
The primary urban surveillance network, described by Chinese state media as "the largest video surveillance system in the world." SkyNet covers streets, transport interchanges, and public spaces across major cities, and is the foundation layer on which facial recognition and other AI analytics run.
Programme 02 · Rural
雪亮工程Sharp Eyes, Residential and rural coverage
Extends surveillance into villages, residential compounds, and rural areas where SkyNet's urban focus doesn't reach. Uniquely, Sharp Eyes is designed to encourage citizens to monitor camera feeds from home, via mobile apps and TV set-top boxes, turning everyday viewers into an auxiliary surveillance layer.
Programme 03 · Data fusion
Blacklist integration, merging video, financial, and regulatory data
Camera feeds combine with court judgments, financial records (via mandatory Alipay / WeChat Pay), and regulatory agency data to populate national and regional blacklist databases. These blacklists, not a single score, are what drive the real-world penalties: flight and rail ticket bans, blocked loan applications, and restricted access to government services.

Variables, units, and source classes.

Each record in the 100-city index documents the following ten variables.
Variable names match the JSON fields used by the JavaScript renderer.

Variable Unit / type Definition Source class
city, country, region text Standard place identifiers. Region grouping follows UN M.49. n/a
cams1k cameras / 1,000 residents Government-operated or government-accessible CCTV cameras per 1,000 residents, as defined by the source studies (Comparitech / NeoMam / Surfshark methodology). Excludes purely private commercial cameras and household systems. Industry / NGO
camskm cameras / km² Same camera count divided by city land area. Useful when comparing dense cities (e.g. Seoul, Beijing) against sprawling ones (e.g. Sydney, Los Angeles). Industry / NGO
camsD % change vs 2015 Percentage change in government-operated CCTV stock between 2015 and most recent available year (typically 2024 or 2024/25). Country-level estimate, with city overrides where firm city-level data exists. Industry / NGO
crimeD % change vs 2015 Percentage change in total police-recorded offences over the same reference window. Null where no comparable time series exists (state-controlled jurisdictions). Selection rules in § 3.3. National statistics
trustD % change vs 2015 Relative change in the proportion of citizens reporting they are not concerned about state data handling. Pew (US, Anglosphere), Special Eurobarometer 487a then 551 (EU), Edelman Trust Barometer (rest of world). Selection rules in § 3.4. Survey houses
gdpr text Applicable national or sub-national privacy framework, as of 2026.Q1. Regulatory

How the 100 cities were selected, how each variable is constructed, and what the limitations are.

Sections 3.1 through 3.8 below define every construction rule, indicator-selection convention, and acknowledged limitation in this dataset. Designed so a fact-checker can audit any cell on the page.

3.1Sample frame

The 100 cities are inherited from the Comparitech / NeoMam global CCTV index (2025 update), which orders cities by government-operated camera density per 1,000 residents. We did not re-sample; the rank ordering is theirs. The frame favours cities for which credible enumeration is possible, which means it under-represents jurisdictions with no public CCTV statistics, parts of Central Asia, Sub-Saharan Africa, and the smaller Pacific states. Coverage by region: Asia 31, Europe 33, Americas 19, Middle East 9, Africa 11, Oceania 2.

3.2Camera-count definitions

Camera counts are sourced from Comparitech 2025 and IHS Markit / Omdia industry reports, cross-referenced with national security-industry associations (BSIA for the UK, SSAIB, ESS Industry Group for Germany).

Included: cameras operated by, or routinely accessible to, government authorities, typically municipal police, transit police, federal law-enforcement agencies, and integrated "Safe City" platforms with formal data-sharing agreements.

Excluded from the headline figure: residential doorbell cameras, private commercial CCTV with no police-access agreement, household security systems.

Chinese cities are reported using the source-study definition, which counts only formally classified state cameras. The actual penetration in Chinese cities is materially higher; including private and on-request-accessible cameras would push China's largest cities to the top of the index. We publish the lower, narrower figure because (a) it is the figure on which the source studies converge and (b) it preserves comparability with non-Chinese rows. A separate methodology note in the main report discusses the higher estimate.

3.3Crime indicator selection

For each country, the crime delta uses the highest-coverage publicly-released crime series available from the national statistical office or police:

Recording-rate caveat.
Police-recorded crime is affected by reporting rates and recording-practice changes. HMICFRS independently assessed UK police compliance with recording standards as rising from 80.5% in 2014 to 94.8% in recent reports; we cite this caveat in any single-jurisdiction analysis. Victim-survey series (CSEW, Eurostat ad-hoc modules) often tell a different and more favourable story for property crime; we treat the divergence between the two as a finding to surface, not an error to suppress.

3.4Trust indicator selection

The two backbone series for public concern about state data handling are:

Country breakdowns are available from both surveys where sample size permits; we use national readings where present, otherwise the EU or regional aggregate, badged reg. For non-democratic jurisdictions where state-controlled survey data lacks independent oversight, the cell is marked ? and no direction is published. Supplementary regional series used: Edelman Trust Barometer (28 markets, annual), Latinobarómetro (Latin America), Afrobarometer (selected African states).

3.5Scope taxonomy

Each delta cell carries a scope badge marking the geographic level at which the underlying series is available:

3.6Limitations and known biases

3.7Reproducibility and updates

Dataset version 2026.1. The cities array, country defaults and city overrides are present in the source HTML of this report and can be extracted directly for verification. Re-runs against newer source releases are produced annually. Corrections to specific cells are welcome via [email protected] and are reflected in versioned changelogs published alongside subsequent releases.

Suggested citation SpyShop Europe Research (2026). 2026 City Surveillance Index, Cameras, Crime, and Public Trust across 100 cities. Dataset v. 2026.1. Reference period 2015–2024/25. Available at spyshop.eu/research.

Named sources, organised by data axis.

Every figure traces to one of the publishers below.
URLs link to the primary release. Replace any figure with project-dataset values and the citation chain is already in place.

Camera density (cams1k, camskm)
  • Comparitech, The Most Surveilled Cities in the World, 2025 update. comparitech.com
  • NeoMam Studios, global CCTV city-ranking studies, 2024–2025 releases.
  • IHS Markit / Omdia, Video Surveillance Intelligence Service, annual market and installed-base reports.
  • Surfshark, Surveillance State Report 2024. surfshark.com
  • British Security Industry Association (BSIA), UK national CCTV stock estimates, 2011 / 2020 / current.
  • Grand View Research, Video Surveillance Market Size Report, $148bn-by-2030 forecast.
  • Mordor Intelligence, regional video-surveillance market reports.
Recorded crime (crimeΔ)
  • Eurostat, Crime and criminal justice database, online code crim_off_cat. ec.europa.eu/eurostat/web/crime
  • UK, MOPAC, Mayor's Office for Policing and Crime, London recorded-offence releases.
  • UK, HMICFRS, Her Majesty's Inspectorate of Constabulary, recording-quality assessments. hmicfrs.justiceinspectorates.gov.uk
  • UK, ONS, Crime Survey for England and Wales (CSEW) victim-survey series.
  • Germany, BKA, Bundeskriminalamt, Polizeiliche Kriminalstatistik (PKS) 2024. bka.de
  • Germany, Polizei Berlin, Kriminalitätslagebild Berlin 2024.
  • France, SSMSI, Service Statistique Ministériel de la Sécurité Intérieure, Interstats. interieur.gouv.fr/Interstats
  • Sweden, BRÅ, Brottsförebyggande rådet, official crime statistics. bra.se
  • USA, FBI, Uniform Crime Reporting (UCR) and NIBRS. fbi.gov/services/cjis/ucr
Public trust / data-protection concern (trustΔ)
  • Pew Research Center, Americans and Privacy series, 2019 and 2023 readings. pewresearch.org
  • Special Eurobarometer 487a, The General Data Protection Regulation, 2019. europa.eu/eurobarometer
  • Special Eurobarometer 551, The Digital Decade, 2024. europa.eu/eurobarometer
  • Edelman Trust Barometer, annual, 28 markets. edelman.com
  • Latinobarómetro, Latin America trust series.
  • Afrobarometer, selected African states.
Academic literature
  • Piza, E. L., Welsh, B. C., Farrington, D. P., Thomas, A. L. (2019). "CCTV surveillance for crime prevention: A 40-year systematic review with meta-analysis." Criminology & Public Policy, 18(1), 135–159., the 13%-effect meta-analysis referenced throughout this report.
  • Welsh, B. C. & Farrington, D. P. (2009). Making Public Places Safer: Surveillance and Crime Prevention. Oxford University Press., predecessor meta-analysis.
Regulatory frameworks (gdpr column)
  • GDPR (EU), Regulation (EU) 2016/679. gdpr.eu
  • UK GDPR / DPA 2018, Information Commissioner's Office. ico.org.uk
  • CCPA / CPRA (California), California Office of the Attorney General. oag.ca.gov
  • PIPL (China), Personal Information Protection Law of the People's Republic of China.
  • DPDP Act 2023 (India), Ministry of Electronics and Information Technology.
  • APPI (Japan), Personal Information Protection Commission. ppc.go.jp
  • PDPA (Singapore), Personal Data Protection Commission. pdpc.gov.sg
  • Full list of 30+ frameworks referenced per city row: see the Privacy Law column of the main index.