BRENNEN

Industries / Finance and business services / Security guard services

Training and licensing

This is the work of teaching guards the job and the law and testing them, and the state checks and licences that decide who may work as a guard.

4

Milestones achieved, from 1971 to 2021.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1971

    Little or no training

    James Kakalik and Sorrel Wildhorn of RAND ran a 16-month study of private police in the US for the Department of Justice, including their "licensing and regulatory standards". A 2010 review for the Bureau of Justice Statistics reports a RAND finding "that 65% of private security personnel had received no training before beginning their job assignments", and notes that researchers, citing Kakalik and Wildhorn, recommended that licensing and regulation "start at the state level".

    rand.orgojp.gov

    Credit

    James S. Kakalik and Sorrel Wildhorn, RAND

  2. 2004

    Licences in Britain

    The Private Security Industry Act 2001, given Royal Assent on 11 May 2001, created the Security Industry Authority (SIA) and made it "an offence to work in those sectors without a licence", with up to six months in prison, a fine of up to £5,000, or both. The SIA began issuing licences in 2004. By the end of May 2008 it had issued over 248,000, and the National Audit Office put compliance at "over 90%", though the SIA had needed "an additional £17.4 million of public funding".

    legislation.gov.uknao.org.ukpersonneltoday.com

    Credit

    UK Parliament; Security Industry Authority

  3. 2007

    Licensed without right to work

    In November 2007 the Home Office confirmed that "About 5,000 illegal immigrants were mistakenly cleared to work as security staff", and moved to check the immigration status of 40,000 licence holders. The SIA's position was that it is "employers' responsibility to check whether staff are entitled to work in the country".

    LessonA licence proves only what the licensing body actually checks.

    personneltoday.com

    Credit

    Security Industry Authority and Home Office (a failure)

  4. 2021

    Top-up training after Manchester

    The Manchester Arena Inquiry heard that one witness's licence training had no in-person part and that in another case candidates were told the exam answers, complaints the chair did not investigate; it found that training by employers after the licence was voluntary and "often online". From 1 October 2021 the SIA required door supervisors and security guards to hold a first aid qualification and to take top-up training, with "updated counter-terror training and advice for emergencies and incidents", before renewing a licence.

    gov.ukassets.publishing.service.gov.uk

    Credit

    Security Industry Authority; the Manchester Arena Inquiry

Watch for

  1. California adds training hours

    California's SB 1203 was chaptered, or written into the state's statutes, on 30 September 2026, by CalMatters' bill tracker. As the law firm Ogletree described it before it passed, it raises the 32 hours of skills training that a new guard must complete within six months to 42 hours from 1 July 2028, requires in-person classroom teaching including eight hours of role-play practice in calming people down, and raises the yearly refresher from eight hours; the sources disagree on whether the new figure is 12 or 16 hours.

    Achieved whenThe requirements take effect and the state's Bureau of Security and Investigative Services publishes course rules under them.

    calmatters.digitaldemocracy.orgogletree.com

    Claimed by

    California Legislature, September 2026

Open

  1. 01

    A world table of licences

    A free table of the rules for working as a guard in each country, state and province: who issues the licence, training hours before the first shift, how much must be in person, the exam, the yearly refresher, background and right-to-work checks, and the extra rules for armed guards, each with a link to the statute or regulator page and the date it was checked.

    The brief
    Credit · unclaimed
  2. 02

    Test AI role-play trainers

    A public test of AI role-play trainers, in which a trainee talks to an AI voice playing a member of the public. It would hold 40 scenarios drawn from inquiry findings (a member of the public reporting a suspicious person, starting an evacuation, a lost child at an event, calling for first aid), a marking guide written by three licensed trainers, and a measure of how often the AI's marks agree with the trainers'.

    The brief
    Credit · unclaimed

Open 01 · the brief

A world table of licences

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A free table of the rules for working as a guard in each country, state and province: who issues the licence, training hours before the first shift, how much must be in person, the exam, the yearly refresher, background and right-to-work checks, and the extra rules for armed guards, each with a link to the statute or regulator page and the date it was checked.

Why it is still open

The 2010 review for the Bureau of Justice Statistics found that US states "vary greatly in their requirements" and that "All but nine states" license guard companies or guards in some way. The fullest current US list found, Belfry Software's state-by-state guide updated in November 2026, is a vendor's blog post that links agency pages, cites no statutes and covers no other country. Rules keep changing, as California's SB 1203 shows. No sourced worldwide table was found.

ojp.govbelfrysoftware.comcalmatters.digitaldemocracy.org

Built with what exists

A web agent to collect statutes and regulator pages, such as Browser Use 1.0 (see the Note "Deleting 80% to start over"); an AI model that reads long legal texts in many languages, such as Claude Opus 5.5, to draft each entry for a person to check; a spreadsheet and a static website.

  1. First step

    Code 10 places from their own statute pages, among them California before and after SB 1203, England and Wales, Ontario and South Africa.

  2. Done looks like

    All 50 US states and the District of Columbia, Canada's provinces and territories, Australia's states and territories, the 27 EU countries, the UK, India, South Africa and at least 20 more countries; two people code each entry separately and agree on at least 95% of fields; the table is checked again every quarter.

Open 02 · the brief

Test AI role-play trainers

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public test of AI role-play trainers, in which a trainee talks to an AI voice playing a member of the public. It would hold 40 scenarios drawn from inquiry findings (a member of the public reporting a suspicious person, starting an evacuation, a lost child at an event, calling for first aid), a marking guide written by three licensed trainers, and a measure of how often the AI's marks agree with the trainers'.

Why it is still open

Exec sells AI voice practice in which "Security officers practice verbal de-escalation for confrontational encounters", and its page gives no outcome data. California's SB 1203 will require eight hours of in-person role-play. The Manchester Arena Inquiry asked for checks that online training "is carried out and that information is absorbed and understood". No published test of AI role-play training for guards was found.

exec.comogletree.comassets.publishing.service.gov.uk

Built with what exists

Speech-to-text such as ElevenLabs Scribe v3 (see the Note "Fewer word errors"); a synthetic voice and an AI model such as Claude Opus 5.5 to play the other party and mark each run; a spreadsheet for the trainers' marks.

  1. First step

    Write five scenarios from the Manchester Arena Inquiry's findings and have one licensed trainer write their marking guide.

  2. Done looks like

    40 scenarios; 200 practice runs by adult volunteers who agree to take part, marked by two trainers and by the AI; the rate of agreement published, with a target above 85%; recordings deleted after marking, and no volunteer graded or named.