Industries / Finance and business services / Security guard services
Reports
This is the work of recording a shift: proof that rounds were walked, logs of visitors and events, incident reports for clients, insurers and police, and the handover to the next shift.
4
Milestones achieved, from 1878 to 2022.
1
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1878
A clock that proves rounds
A watchman's clock recorded rounds on paper. "Keys, each with a different shape, would be hung at each checkpoint along the guard route", and turning a key in the clock embossed its mark on a paper chart, so a supervisor could see when each point was visited. Abraham A. Newman began making stationary watchclocks in 1878; his business became the Newman Watchclock Corporation in 1923 and then Detex, which "discontinued the manufacture of mechanical watchclock systems on Dec. 31, 2011".
Credit
Abraham A. Newman; Detex
2009
Certificates on file, expired
The Government Accountability Office (GAO) checked 663 randomly chosen records of contract guards at buildings protected by the Federal Protective Service and found that "62 percent of the guards had at least one expired certification". In covert tests, GAO investigators carrying the parts of a homemade bomb "successfully passed undetected through security checkpoints" at all 10 high-security federal buildings tested. In 2010 GAO found that "7 of 7 guard contractors GAO reviewed" were not meeting training and certification requirements.
LessonA guard's training record is only as good as the checks that someone makes against it.
Credit
US Government Accountability Office
2017
A warning goes nowhere
About 15 minutes before a concert at Manchester Arena ended on 22 May 2017, a member of the public told a steward of his concern about a man in the foyer, who turned out to be the bomber. The steward had no radio. A second steward who had one tried to call it in, could not get through because someone else was talking, and "made no further efforts to communicate what he had been told to anyone else".
LessonThe inquiry found the 15 minutes "a sufficient period of time both for an investigation" of the concern "and for decisive action".
Credit
Manchester Arena Inquiry, Volume 1, June 2021; Showsec, the crowd management contractor (a failure)
2022
Guard software consolidates
In June 2022 Trackforce Valiant bought TrackTik, a Canadian maker of "security workforce management" software, for an undisclosed sum, and TrackTik's chief executive said the deal "positions us as the dominant leader for security workforce technology". Guards file incident reports in TrackTik, where Trackforce added an AI report writer in 2025.
Credit
Trackforce Valiant; TrackTik
Watch for
AI rewrites guards' notes
Trackforce says ReportPro AI, inside its TrackTik software, "transforms informal notes into compliance-ready documentation", writes summaries for supervisors, keeps "Original content... stored alongside AI modifications" and marks AI-written text. It gives no figures for time saved or errors, and a customer's statement that report quality "significantly improved" comes with no method.
Achieved whenA guard company or researcher outside Trackforce compares AI-assisted reports with guards' original notes and with what happened, and publishes the error counts.
Claimed by
Trackforce, September 2025
Open
01
Test AI-written incident reports
A public test of AI report writers on guard work. It would hold 200 invented incidents (a slip in a lobby, a door found open, a fire alarm, a lost child, an unattended bag), each with a list of the facts a report must contain, and spoken notes for each read by synthetic voices, so that no real person is recorded. Each tool turns the notes into a report, and a script and two supervisors count facts left out, facts added that were not in the notes, and wrong times.
The briefCredit · unclaimed02
Test that warnings reach someone
A public test of whether reporting software, dispatch software or an AI assistant passes a reported concern to a person who can act, within the time a site's rules allow. It would hold written escalation rules for 10 invented sites and 300 scripted reports, by radio call or app message, each with the right recipient and deadline, and it would score each system on who it told and how fast.
The briefCredit · unclaimed
Open 01 · the brief
Test AI-written incident reports
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 report writers on guard work. It would hold 200 invented incidents (a slip in a lobby, a door found open, a fire alarm, a lost child, an unattended bag), each with a list of the facts a report must contain, and spoken notes for each read by synthetic voices, so that no real person is recorded. Each tool turns the notes into a report, and a script and two supervisors count facts left out, facts added that were not in the notes, and wrong times.
Why it is still open
Trackforce publishes no accuracy figures for ReportPro AI. The only independent tests found are of AI-drafted police reports: a 2024 study by Ian Adams and colleagues found "no time savings" from Axon's Draft One, and in a 2026 study, 92 senior officers rated AI-drafted reports "substantively and significantly worse on ACCURACY". No test of AI report writing on guards' reports was found.
Built with what exists
Speech-to-text such as ElevenLabs Scribe v3 (see the Note "Fewer word errors") and ElevenLabs' synthetic voices; AI models such as Claude Opus 5.5 and the open-weight Mistral Large 4 (see the Note "A trillion open parameters") as baselines; Python for scoring.
First step
Write 20 incidents and their fact lists, record the notes with synthetic voices, and score one tool by hand.
Done looks like
200 incidents with fact lists checked by two supervisors; at least three tools scored, including at least one sold to guard firms; omissions, additions and wrong times published for each tool.
Open 02 · the brief
Test that warnings reach someone
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public test of whether reporting software, dispatch software or an AI assistant passes a reported concern to a person who can act, within the time a site's rules allow. It would hold written escalation rules for 10 invented sites and 300 scripted reports, by radio call or app message, each with the right recipient and deadline, and it would score each system on who it told and how fast.
Why it is still open
At Manchester Arena a concern passed from a member of the public to a steward without a radio, then to a steward whose call was blocked by radio traffic, and went no further, with 15 minutes left that the inquiry judged enough for "decisive action". ReportPro AI writes summaries for supervisors, but no published test of whether any tool gets a warning to the right person in time was found.
Built with what exists
Scripted radio calls voiced with synthetic voices; speech-to-text such as ElevenLabs Scribe v3; AI models such as Claude Opus 5.5 to read the rules and the reports; Python to time and check each routing.
First step
Write escalation rules for one invented arena and 30 scripted reports, and run them through one guard app by hand.
Done looks like
300 reports across 10 sites; at least three systems scored; every report that reached nobody, or reached someone after the deadline, listed.