private-security-workforce

Published by Mercenomics, an independent data archive on the business of force for hire.

How many private security guards five countries have, year by year, from each government's own records, and, where the same official source also counts police, how many guards there are for every police officer.

Country	Private security	Police	Years	Source
United States	Security guards employed	Police and sheriff's patrol officers employed	2014-2025	Bureau of Labor Statistics
Japan	Security guards employed by authorised security companies	not included	2007-2025	National Police Agency
South Africa	Security officers registered with the regulator and recorded as employed	not included	2011-2023	Private Security Industry Regulatory Authority
United Kingdom	People holding an active security licence	not included	2004-2026	Security Industry Authority
Australia	Security officers and guards employed	Police employed	1986-2026	Australian Bureau of Statistics

What it shows

• United States: 1,283,470 security guards in May 2025 against 670,520 police and sheriff's patrol officers, or 1.91 guards for every police officer, up from 1.69 in 2014.
• Japan: 596,985 security guards at the end of 2025, working for 10,949 authorised security companies.
• South Africa: 577,444 security officers were registered and recorded as employed on 31 March 2023, up from 411,109 in March 2011.
• United Kingdom: 460,138 people held an active Security Industry Authority licence on 1 January 2026, up from 242,985 in January 2008.
• Australia: on average 62,800 security officers and guards and 75,300 police were employed in 2025, or 0.83 guards per police officer. Australia has had roughly as many guards as police since the early 2000s, with the ratio moving between about 0.8 and 1.1.

Things to know before comparing countries

The five countries count different things, so compare each country's trend over time, not the size of one country against another.

• United States: an employer survey (Occupational Employment and Wage Statistics). It counts people employed in the occupation in May each year, whether they work for a security company or directly for a shop, hospital or other employer. It does not count the self-employed. The rse_pct column gives the survey's own estimate of its sampling error.
• Japan: the National Police Agency's count of guards employed by companies authorised under the Security Business Act, at 31 December. It covers every kind of guarding, from traffic guards to cash transport, and both regular and temporary staff (split from 2021).
• South Africa: the Private Security Industry Regulatory Authority's count of registered security officers recorded as employed ("active") at the end of its financial year, 31 March. Many more people are registered but not working: 2,816,638 security officers were registered in March 2023, of whom 577,444 were active. Figures for 2013 are missing, and the reports for 2023/24 and 2024/25 could not be read for this version.
• United Kingdom: the number of people holding at least one active licence from the Security Industry Authority. A licence shows who may legally do licensable security work, not who is working at that moment, and some security roles do not need a licence. One person can hold several licences, so licences outnumber licence holders. The Authority issued its first licences in 2004 (466 licence holders in March 2004), so the rise to about 2008 reflects licensing being taken up, not the workforce growing.
• Australia: a household survey (Labour Force Survey), so the quarterly figures move up and down from sampling alone; for example police jump from 64,400 to 84,900 between two November quarters. The annual table averages the four quarters of each year. It counts people by the occupation of their main job.
• Police are included only where the same source counts both guards and police (US and Australia), so that both numbers are measured the same way. For Japan, South Africa and the UK the sources used do not give a matching police headcount.

The data

File	Contents
data/clean/annual_by_country.csv	One comparable figure per country and year, with police and guards per police officer where available (104 rows)
data/clean/security_workforce.csv	Every figure used, one row per country, period and measure, including UK licences by sector, Japan's regular and temporary guards and company counts, and Australia's quarterly estimates (2,534 rows)

Columns of security_workforce.csv:

Column	Meaning
country	Country
period	Year (YYYY), month (YYYY-MM) or date (YYYY-MM-DD) the figure refers to
period_type	year_end, month, mid_quarter_month or date
workforce	private_security or police
measure	What is counted, for example guards_employed, licence_holders, active_licences_door_supervision, security_companies
value	The figure, exactly as published (Australia: summed across sexes and states)
unit	persons, thousand persons, licences or companies
rse_pct	United States only: the survey's relative standard error of the estimate, in percent
source	The agency that published the figure
source_file	The file in data/raw/ or the address it came from
source_location	Where in that file the figure is

Columns of annual_by_country.csv: country, year, reference_point (the date or period behind the figure), private_security, private_security_measure, police, police_measure, guards_per_police_officer, note.

How the numbers were checked

Run by code/03_check.py; results in data/clean/check_report.md.

1. Japan: twelve years appear in more than one official publication (three police white papers and the 2025 report). Every one agrees. Regular plus temporary guards equals the total in every year they are split.
2. United States: for 2016 to 2023 each figure was read twice, from the Bureau's web tables and from its spreadsheets. All 16 match.
3. Australia: the figures summed from the Bureau's detailed rows match its own summary table for all 80 November-quarter figures checked.
4. United Kingdom: the monthly series has no missing months, and where the Authority's 2021 and 2026 releases overlap they differ by less than 0.1%. The sector columns add up to the total in every month except April 2018, where the source itself differs by one licence; it is kept as published.
5. South Africa: ten figures appear again in other PSIRA reports or presentations, covering eight financial years, and all match. Summing the nine provinces in the 2021/22 report gives the national total for 2019/20 to 2021/22 exactly, but 1,101 fewer than the national table for 2018/19; the national table is used.
6. No figure appears twice, and nothing has been estimated or filled in.

Reproduce it

Needs Python 3.

pip install -r requirements.txt
python code/01_download.py   # UK, Japan and Australia files; prints the US file addresses
python code/02_build.py
python code/03_check.py
python code/04_figure.py

The Australian file (56 MB) is not stored in the repository; 01_download.py fetches it, and data/raw/australia/eq08_protective_service_extract.csv keeps the rows used. The US Bureau of Labor Statistics blocks automated downloads, so its figures were read from its official files in a browser and stored in data/raw/us/ with the address of each file.

Sources and license

Sources, download dates and terms are in data/README.md. Code: MIT (see LICENSE). The clean dataset: CC BY 4.0. Source material keeps its own terms: UK figures contain public sector information licensed under the Open Government Licence v3.0; Australian figures are from the Australian Bureau of Statistics under CC BY 4.0; South African figures are taken from PSIRA annual reports as tabled in Parliament (only the figures are used; the reports are not redistributed); Japanese figures are edited from National Police Agency publications (「警備業の概況」「警察白書」（警察庁）を加工して作成) under the Public Data License 1.0 and are not an official National Police Agency product; US figures are public-domain works of the US government.

Cite

Azom, Kazi Affnann (2026). private-security-workforce: official counts of private security guards by country, with police where the same source counts both (Version v1.0.0) [Data set]. Mercenomics. https://doi.org/10.5281/zenodo.23768202

• This version (v1.0.0): 10.5281/zenodo.23768202
• All versions (always points to the latest): 10.5281/zenodo.23768201

See also CITATION.cff.