Why guard work stays with people
Most of a guard’s shift happens in a building, not on a screen. Patrolling a site, checking credentials at an entrance, responding to an alarm and calming down an argument are physical, local jobs. Software can flag that a door opened at 2 a.m. Someone still has to walk there, look, and decide what it means.
The second reason is accountability. Guards write incident reports, hand them to managers, insurers and sometimes police, and may be asked about them later. A camera system can log motion. It cannot stand behind a judgment call about whether a person was trespassing, lost, or in trouble. That mix of presence and responsibility is why the task split above leans the way it does.
Deterrence matters too. Part of the value of a uniformed guard is that people behave differently when one is nearby. A sensor does not change behavior the same way, which is why sites that add analytics often keep the posts.
What AI does, what it helps with, and what stays with people
Start with the measured share AI can handle today: 14 out of 100 on our coverage scale, which asks what share of task time AI can do now. The pattern is simple. Watching scales. Showing up does not. Our coverage method explains how that share is built from task time.
The work AI can take outright is narrow: 9%. It is the repetitive screen work, such as scanning many camera feeds for movement and sorting routine access logs, where a model never gets bored at hour seven.
Assisted work is bigger: 14%. Here the software triages and the guard acts. Video analytics narrow dozens of alerts down to a handful worth a walk. Guard tour apps time-stamp checkpoints and prompt the next round. Report tools draft the first version of an incident write-up, which the guard then corrects, because the details that matter came from standing there.
The rest stays with a person: 77%. Foot patrols of a parking structure, searching a bag, escorting an employee out after a termination, giving first aid, directing people in a fire alarm, and testifying about what happened. None of that is a prediction problem.
How strong is the evidence on guards?
Honest answer: thin. Our quality score carries an evidence grade of D, which means no study has yet tested an AI or robot system head-to-head against a qualified guard on this job’s real duties. So we publish no parity number for security guards, and you should treat any site that gives you one as guessing. Our quality parity method sets out what counts as a test.
What would settle it is specific: a trial on live sites, comparing detection accuracy, false alarms, response time and incident outcomes between a staffed post and a sensor-plus-remote-monitoring setup, with the results published. Vendor case studies from one customer do not clear that bar.
The labor data is firmer. The Bureau of Labor Statistics counted about 1,283,470 US security guard jobs with median pay of $38,020 a year, and projects employment roughly flat, up about 0.9% from 2025 to 2035 (BLS, 2025). Flat is not shrinking. It also means fewer new openings than a growing field, so entry-level competition matters.
When the mix could shift
Most likely after 2042 (8 in 10 of our scenarios). We explain what that window does and does not mean on the replacement year method page.
Two things could pull it earlier. First, cheaper mobile robots: most of this job’s exposure is physical, and the robotics tier above is mobile ground machines, not fixed arms, so progress on wheeled and legged patrol units matters more here than progress in chatbots. Second, remote monitoring centers, where one operator watches many sites and only dispatches a human when needed, which can thin out overnight posts first.
Two things hold it back. Buildings are messy: stairs, weather, elevators, loading docks and crowds break robots that work fine in a demo. And the total cost of a sensor-and-robot stack, shown in the cost panel above, still has to cover downtime, maintenance and a human on call, which narrows the saving on many sites.
What to do: if your post is pure camera watching, ask to be trained on response, access control and reporting, which are the parts the task list above keeps with people.
How to stay needed in security work
Lean into the duties that need a body and a decision. Physical response to alarms and incidents. Access control at the door, including the judgment calls about who gets in. Clear, defensible incident reporting that holds up weeks later.
Two skills raise your value fast. One is de-escalation and conflict handling, which is the part of the job that keeps sites out of court. The other is working the tools: video analytics dashboards, guard tour software and access systems, so you can verify an alert instead of chasing every one.
If you want to move sideways, closer roles are worth a look. Compare the task mix for Gambling Surveillance Officers and Gambling Investigators, Transportation Security Screeners and First-Line Supervisors of Security Workers, or put any two jobs next to each other with our side-by-side comparison. The wider other protective service workers family and the administrative support sector page show where guard work sits among neighboring roles, and our list of jobs that mostly need a person shows which duties tend to hold up.
Our headline figure for this job, 80 out of 100 (higher is safer), comes from open data: O*NET task time, BLS employment and pay, and published research. The full scoring methodology shows every input, and you can see how other roles compare in the full job rankings. If you work a post that these numbers do not describe well, the patrol and response tasks above are the ones to check first.