Emergency Department Security: The 2026 AI Physical Security Sector Playbook for ED Operations, Behavioral-Health Boarding, and the Violence-and-Falls Convergence
The emergency department is where hospital violence, falls, and behavioral-health boarding converge on the same patients in the same hallways. This sector playbook maps the ED security environment and shows where AI detection on existing cameras closes the gaps staffing cannot.
No department in an American hospital absorbs more violence, more crowding, and more unwitnessed falls than the emergency department, and the three are not separate problems. They converge on the same patients, in the same hours, in the same hallways. This sector playbook treats emergency department security as the operational discipline it has become, and shows where AI physical security earns its place inside it.
The emergency department is the one part of a hospital that cannot turn anyone away, cannot control its own census, and cannot schedule its workload. Under the Emergency Medical Treatment and Labor Act, every person who arrives must be screened and stabilized regardless of insurance, behavior, or intoxication. That legal duty makes the emergency department the highest-acuity, lowest-predictability environment in American health care, and it is precisely that combination that turns the emergency department into the hospital's center of gravity for physical-security risk. Violence concentrates here. Falls concentrate here. Elopement, behavioral crises, and boarding all concentrate here. And the staff who absorb that risk are clinicians trained to treat patients, not to run a security operation.
This is a sector playbook for the people who own that risk: emergency department directors, hospital security and safety leaders, chief nursing officers, and the risk managers who answer for what happens in the waiting room and the hallway. It is not a clinical protocol and it is not a pitch to put guards on every door. It is a structured look at why the emergency department is a distinct security environment, what the primary-source data actually says about the violence-and-falls convergence, and where AI computer-vision detection layered onto the cameras a hospital already owns closes gaps that staffing alone cannot. The orientation is the same one IntelliSee brings to its broader healthcare workplace-violence detection playbook: detection answers what is happening, not who it is happening to.
Why the emergency department is its own security environment
Most hospital security planning treats the building as one envelope with a single perimeter and a uniform interior. The emergency department breaks that model. It is the only 24-hour open door in the facility, it runs a public waiting room directly adjacent to a clinical care area, and it holds a patient population selected for exactly the traits that drive incidents: acute pain, intoxication, psychiatric crisis, withdrawal, dementia, grief, and the frustration of long waits. A general medical-surgical floor sees a screened, admitted, oriented population. The emergency department sees everyone, in their worst hour, before anyone has been assessed.
The structural features compound one another. Crowding is chronic because boarding admitted patients in emergency department hallways is now routine, which the Agency for Healthcare Research and Quality convened a national summit to address in October 2024 after describing emergency department boarding as a public-health crisis. Crowding lengthens waits, and long waits are themselves a documented trigger for violence and for patients leaving against medical advice. Crowding also splits staff attention across more patients than the physical space was designed for, which is the condition under which an unwitnessed fall in a hallway bed can go undiscovered. Every one of these is a security variable, and every one of them is worse in the emergency department than anywhere else in the building.
An open door, a public waiting room, and a population no one has assessed yet
The emergency department is the only part of the hospital that combines unrestricted public access, a clinical care area feet away from an unscreened waiting room, and a patient mix defined by intoxication, psychiatric crisis, and acute distress. A medical-surgical unit can lock down. An operating suite controls every entry. The emergency department, by federal mandate, cannot. That is why security strategies built for the rest of the hospital fail at the emergency department door, and why the emergency department deserves a sector playbook of its own rather than a paragraph in a hospital-wide plan.
What the violence data actually says
The claim that emergency department staff face extraordinary violence is not anecdotal. It is one of the best-documented occupational-risk findings in American health care, and it rests on three independent primary sources that point the same direction. The first is the U.S. Bureau of Labor Statistics, whose injury data the Government Accountability Office analyzed in its 2016 report on health-care workplace violence. The GAO found that private-sector health-care workers in inpatient facilities experienced workplace-violence injuries requiring days away from work at a rate at least five times higher than private-sector workers overall, and that health care accounts for roughly 73 percent of all nonfatal workplace-violence injuries in the economy. Health care is the single most violent civilian work setting in the United States, and within health care, the emergency department is the most exposed point.
The second source is the American College of Emergency Physicians. In its January 2024 national poll, 91 percent of emergency physicians reported that they or a colleague had been a victim of violence in the past year, and 89 percent agreed that violence in the emergency department has harmed patient care, up from 77 percent in 2018. The third is the Emergency Nurses Association, whose 2024 member survey found that 56 percent of responding emergency nurses had been physically assaulted, verbally assaulted, or threatened with violence in the previous 30 days. A separate line of ENA and ACEP survey work has long found that roughly 70 percent of emergency nurses report being hit or kicked on the job. These are not outliers. They are the consistent finding across the two professional bodies that represent the people staffing the room.
What makes the data operationally useful is not the headline percentages but the pattern underneath them. Violence in the emergency department is concentrated at predictable points: the triage window, where the first contact with an agitated or intoxicated person happens; the waiting room, where the wait itself is the trigger; and the behavioral-health and boarding areas, where psychiatric patients are held for hours with nowhere to go. The same survey work that quantifies the violence also documents the institutional failure to respond, with large shares of physical-assault cases drawing no action against the perpetrator and no follow-up from the hospital. The gap is not awareness. It is the speed and certainty of detection and response at the moments the data says matter most.
The falls problem that hides inside the violence problem
Emergency department security conversations focus almost entirely on violence, which means the second major physical-safety exposure in the room gets overlooked: patients fall in the emergency department, and they fall differently than they do anywhere else in the hospital. The emergency department fall population is younger than the inpatient fall population and far more likely to involve alcohol, recreational substances, and acute medical instability. Peer-reviewed work has found that the single most likely moment for an emergency department patient to fall is during mobilization, especially an unassisted trip to the bathroom, and that increased emergency department volume at the time of arrival, a direct measure of crowding, is associated with an elevated risk of falls.
That last finding is the link the two problems share. The same crowding that drives violence also drives falls, because both are functions of split staff attention and patients held longer than the space was built to hold them. A boarding psychiatric patient, an intoxicated patient sobering up on a hallway gurney, and an elderly patient awaiting an inpatient bed are simultaneously the highest violence risk and the highest fall risk in the department. The reimbursement and severity economics of a fall, once it happens, are developed in the IntelliSee ROI framework on the reimbursement cliff of inpatient falls, and the underlying detection mechanism in the technology briefing on how computer vision identifies falls in real time. The point for an emergency department director is narrower and more immediate: the fall and the assault are not competing for attention, they are the same coverage gap wearing two faces.
Three exposures, one coverage gap
Violence, falls, and behavioral-health boarding are usually managed by three different teams with three different budgets. In the emergency department they collapse onto the same patients, in the same hours, in the same hallways. The shared failure point is detection speed at the moment of the event.
- Concentrated at triage, the waiting room, and boarding areas.
- Long waits and crowding are documented triggers.
- Detection at the trigger points is faster than a passing observation.
- Younger, more often intoxicated than the inpatient fall population.
- Most likely during unassisted mobilization to the bathroom.
- Unwitnessed hallway falls drive the long-lie interval and severity.
- Psychiatric patients held with nowhere to transfer them.
- Highest simultaneous violence and elopement risk in the department.
- Hours of holding with intermittent direct observation.
The convergence: the same crowding that lengthens boarding splits the staff attention that would catch a fall and de-escalate a confrontation. One detection layer on the cameras already covering triage, the waiting room, and the boarding hallways addresses all three exposures at the only point they share, which is the second the event begins.
Behavioral-health boarding is the hardest corner of the room
If the emergency department is the hospital's center of security gravity, behavioral-health boarding is the center of the emergency department's. When a patient in psychiatric crisis arrives and no inpatient psychiatric bed is available, the emergency department holds them, sometimes for hours and frequently for more than a day. The American College of Emergency Physicians and the Agency for Healthcare Research and Quality describe boarding times that average seven to eleven hours and routinely exceed 24 hours when a transfer to an outside facility is required. Published chart reviews have found average psychiatric boarding intervals above 23 hours with total emergency department lengths of stay above 30 hours. During those hours, a patient who may be agitated, suicidal, or disoriented is held in a general environment that was never designed for psychiatric holding.
This is the single highest-risk situation in the department on every axis at once. It is the highest violence risk, because the patient is in crisis and the wait compounds it. It is among the highest fall and self-harm risks, because the patient may be intoxicated, withdrawing, or medicated and is held for an extended period with intermittent observation. And it is the highest elopement risk, because a patient who wants to leave has hours to find an unwatched exit. Continuous one-to-one human observation is the clinical ideal and is rarely fully staffable across every boarding patient for the full duration. This is exactly the gap a detection layer is suited to, not as a replacement for the sitter or clinician, but as the always-on second set of eyes on the exits and holding area that never blinks and never gets pulled to another patient.
Why boarding concentrates every exposure into one patient
A boarding psychiatric patient is simultaneously the department's highest violence risk, highest elopement risk, and among its highest fall and self-harm risks, held for an average of seven to eleven hours in a space built for rapid throughput, not extended holding. Human one-to-one observation is the right standard and is structurally hard to sustain for every boarding patient across a full shift. Posture-and-motion detection on the holding area and the adjacent exits does not replace the sitter. It covers the seconds the sitter looks away, flags a person down or a patient at a controlled door, and routes the alert into the same workflow staff already use, which is the only realistic way to extend coverage without staff who do not exist in the labor market.
Where AI detection fits against the existing security stack
The most common and most reasonable objection from an emergency department director is that the department already has security: badge access on the clinical doors, panic buttons at triage, a guard or two, cameras feeding a recorder, and metal detection in some facilities. None of that is wrong, and none of it is what a detection layer replaces. The honest comparison is about which point in the timeline each modality acts on, and whether anything in the existing stack acts on the event itself in the seconds it is happening rather than before it or after it.
Where Each Layer Acts in the Emergency Department
| Layer | What it does | Where it wins | Where it leaves a gap |
|---|---|---|---|
| Badge access on clinical doors | Restricts entry from the waiting room into the treatment area. | Keeps the unscreened public out of the clinical space. | Does nothing inside the waiting room or the treatment area where most incidents actually occur. Tailgating defeats it. |
| Panic buttons and duress alarms | Staff-activated alert when a person feels threatened. | Fast escalation once a clinician decides to push it. | Depends on a staff member being able to reach and press it during an assault. Silent on falls and elopement entirely. |
| Security officers | Human presence, response, and de-escalation. | Judgment, physical response, and a visible deterrent. | Cannot watch every camera, every hallway, and every boarding patient at once. Coverage is wherever the officer is standing. |
| Recorded CCTV | Captures footage for review after an incident. | Evidence, investigation, and the after-action record. | Passive by definition. No one is watching the wall of monitors in real time, so it documents the event rather than catching it. |
| Weapons / metal detection at entry | Screens for weapons at the door, where deployed. | Intercepts a weapon before it enters the department. | Single chokepoint. No coverage of falls, behavioral escalation, or elopement once the patient is inside. |
| AI computer-vision detection | Existing ED cameras analyzed on-premises for the posture and motion signatures of a person down, a fall, and a person at a controlled exit. No facial recognition, no stored video, no PHI. | Acts on the event itself in real time, across triage, the waiting room, and boarding hallways at once, and routes an alert within seconds into the workflow staff already use. | Not a substitute for officers, clinical judgment, or one-to-one observation. Pairs with them as the always-on trigger, and is not deployed in exam rooms or other camera-prohibited zones. |
The pattern is consistent with what holds across IntelliSee's detection work and with the threat-window logic developed in the briefing on the response-time window that defines security posture: the existing stack acts before the event or preserves the record after it, while computer vision acts on the interval during the event that nothing else is covering. The detection layer is the earliest possible trigger in the response chain, not a replacement for the people in the chain.
How the detection layer works in an emergency department
The mechanism is deliberately unglamorous, which is what makes it deployable in a live clinical environment. An existing IP camera covering the triage window, the waiting room, an ambulance bay, or a boarding hallway streams into an on-premises detection appliance installed in the hospital's own server room. The appliance runs computer-vision models trained on a small, defined set of physical signatures relevant to the emergency department: a person down in a posture and zone where that is unexpected, a sudden vertical-to-horizontal transition that reads as a fall, and a person present at a controlled exit. When a detection threshold is crossed, the platform generates an alert and routes it through the hospital's existing notification infrastructure to the security console, the charge nurse, and where configured the staff mobile devices, with the camera and zone identified.
Three design choices make this clinically and legally viable. First, detection is posture-and-motion based rather than identity based. The platform performs no facial recognition, which keeps it outside the biometric-privacy frameworks that would otherwise stall the project, a distinction IntelliSee details in its briefing on biometric privacy compliance and the state patchwork. Second, the system is zone-aware: a patient lying on a gurney in a treatment bay is not a fall, and a clinician kneeling to a patient is not a person down, because zones and expected postures are defined and tuned during deployment. Third, processing is on-premises, so patient video never traverses a cloud round trip and an internet outage does not compromise detection, which is the prerequisite for HIPAA review and IT security sign-off. The platform answers what is happening, not who it is happening to, which is the same architectural choice that lets it run on the cameras a hospital already owns. The conditions under which the models hold up, low light, motion blur, and partial occlusion in a busy room, are the subject of the briefing on how computer vision handles occlusion, low light, and adversarial conditions.
Why posture detection clears the review that blocks identity-based surveillance
The IntelliSee platform performs object, posture, and motion-pattern detection. It does not perform facial recognition. It does not store video. It does not collect protected health information. In an emergency department, where HIPAA, state biometric-privacy statutes, EMTALA obligations, and patient-dignity expectations all apply at once, this architectural choice is the prerequisite that makes a cameras-on-the-public-areas deployment feasible at all. Detection answers what is happening, a person on the ground in the waiting room or a patient at a controlled exit, rather than who it is. That single design decision is what lets a hospital deploy on the cameras it already owns covering triage, the waiting room, and boarding hallways without triggering the privacy-review cascade an identity-tracking system would require, and it is what survives the deposition question of how the platform handles patient data: it collects no biometric data and stores no video.
Where the case is strongest inside the department
The emergency department is not uniform, and neither is the detection case. The zones below are where the violence-and-falls convergence is sharpest and where existing cameras most often already exist but sit unwatched, which is where a detection layer earns priority placement.
Triage and Intake
The first point of contact with an unscreened, often agitated or intoxicated person, and a documented concentration point for assaults on staff. Detection on the triage camera flags a person down or an escalating situation the moment it begins, rather than relying on the triage nurse to also be a security operator.
The Waiting Room
Where the wait itself is the trigger and where an unwitnessed collapse among a crowded, unmonitored population can sit undiscovered. A person-down detection on the existing waiting-room camera turns a passive recorder into an active alert into the same workflow as a clinical alarm.
Behavioral-Health and Boarding Areas
The highest combined violence, fall, and elopement risk in the department, where patients are held for hours with intermittent observation. Posture-and-motion detection covers the holding area and adjacent exits continuously, extending the sitter rather than replacing them, which is the same augmentation logic developed for continuous monitoring of high-risk individuals.
Hallway Beds and Corridors
Boarding has pushed admitted patients into emergency department hallways, where staff attention is thinnest and an unassisted fall to the bathroom is most likely. Corridor detection pulls these falls into alert routing rather than relying on a passerby to notice a patient on the floor.
Ambulance Bays and Entrances
The open, 24-hour access points where the department is most exposed to outside threats and where a person down outside the door is easy to miss. Detection on the existing exterior cameras extends coverage to the approaches the building cannot lock.
Controlled Exits and Egress
The doors a boarding or behavioral-health patient might use to elope, which a busy clinical team cannot watch continuously. Detection flags a person present at a controlled exit and routes the alert before the patient is gone, a connection that ties the emergency department to the broader facility intrusion and egress coverage picture.
How detection supports the standards an emergency department already answers to
An emergency department director does not get to deploy security technology in a vacuum; it has to map to the standards the hospital is surveyed against. Two are directly relevant. The Joint Commission's workplace-violence-prevention requirements, in effect since 2022, require accredited hospitals to operate a documented workplace-violence-prevention program with worksite analysis, incident tracking, and follow-up, the specifics of which IntelliSee covers in its briefing on the Joint Commission 2026 workplace-violence standards. A detection layer contributes to that program in a concrete way: it produces a time-stamped, location-specific record of incidents and responses that strengthens the worksite analysis and the incident-tracking the standard requires, and it demonstrates an engineering control rather than a policy alone.
The second is the trajectory of state law. A growing number of states have enacted or are advancing health-care workplace-violence-prevention mandates that require risk assessment, prevention planning, and in some cases specific environmental controls, a patchwork IntelliSee tracks in its work on the workplace-violence-prevention-plan mandates emerging from California SB 553 and beyond. For an emergency department, the practical effect is that a documented, always-on detection capability is increasingly not just a safety investment but a compliance artifact, evidence that the hospital has implemented a real environmental control at the point of highest risk rather than a binder of policies. That time-stamped record is the kind of objective documentation surveyors and plaintiffs' attorneys both look for, and that a passive recorder reviewed only after an incident cannot provide.
A note on staffing: augmentation, not substitution
The most damaging way to frame an emergency department detection deployment internally is as a way to cut security officers or nursing staff. It is not, and it should not be sold that way to clinical leadership, the security team, or regulators. Emergency department staffing is constrained by national labor supply, by reimbursement, and by the department's inability to control its own volume. A hospital cannot close the coverage gap by adding people, because the people do not exist in the labor market at the numbers that would watch every hallway, every boarding patient, and every exit at once. What detection does is make the existing team faster and better-positioned: officers and nurses respond to a precise location with a time stamp the moment an event is detected, instead of finding it on a round or by chance. The detection layer sits upstream of the existing response protocol as the earliest trigger, the same structure IntelliSee develops across its healthcare and broader hospital safety work.
What an emergency department deployment actually involves
A sector playbook has to survive the implementation question, because a capability that requires ripping out the camera fleet is not deployable in a live emergency department. The IntelliSee model layers onto existing infrastructure. The platform connects to the hospital's existing IP camera network through its current video management system, and detection runs on a dedicated rack-mounted appliance in the hospital's own server room rather than in the cloud. Most existing emergency department cameras stay in place. Alerts route through the communication infrastructure staff already use, the security console, charge-nurse stations, and mobile devices, so detection appears inside the existing workflow rather than as a separate screen nobody watches. A typical deployment reaches initial coverage of triage, the waiting room, and the priority hallways within days of appliance installation, followed by a tuning period during which detection zones are calibrated to the department's layout and false-positive thresholds are adjusted per camera. The cameras, the video, and the retention policy all stay the hospital's. To model the convergence framework against a specific department's incident log and layout, hospitals can request a risk assessment.
Frequently asked questions about emergency department security
Why is the emergency department considered the highest-risk department for violence?
Because it combines unrestricted public access, a patient population selected for intoxication, psychiatric crisis, and acute distress, and chronic crowding that lengthens waits, which are themselves a documented trigger for violence. The data is consistent across sources: the Government Accountability Office found health-care workers face workplace-violence injury rates at least five times the private-sector average, and health care accounts for roughly 73 percent of all nonfatal workplace-violence injuries. Within health care, the American College of Emergency Physicians reported in January 2024 that 91 percent of emergency physicians said they or a colleague had been a victim of violence in the past year. The emergency department is the most exposed point in the most exposed sector.
How does AI detection handle the fall risk in the emergency department, not just violence?
The same posture-and-motion model that flags an assault or a person down also flags a fall, because a fall is a sudden vertical-to-horizontal transition in a zone where that is unexpected. Emergency department falls are a distinct problem: the population is younger and more often intoxicated than the inpatient fall population, and fall risk rises with crowding. Detection on existing waiting-room and hallway cameras catches an unwitnessed fall and routes the alert within seconds, rather than relying on staff who are split across a crowded department to notice a patient on the floor. In the emergency department, the violence problem and the fall problem are the same coverage gap.
Does camera-based detection in the emergency department violate HIPAA or patient privacy?
Not as IntelliSee implements it. The platform performs posture and motion-pattern detection, not facial recognition. No video is stored or transmitted off the hospital's own network for detection, and no protected health information is collected by the detection layer. Detection answers what is happening, a person on the ground or a patient at a controlled exit, rather than who it is. Cameras are deployed in public and common areas such as triage, the waiting room, ambulance bays, and boarding hallways, not in exam rooms or other camera-prohibited clinical spaces. This architecture is what allows deployment without triggering the biometric-privacy review cascade an identity-based system would require.
How does detection help with behavioral-health boarding specifically?
Behavioral-health boarding concentrates the department's highest violence, fall, and elopement risks into one patient held for an average of seven to eleven hours, often more than 24 hours for transfers. Continuous one-to-one human observation is the clinical standard and is structurally hard to sustain for every boarding patient across a full shift. Posture-and-motion detection on the holding area and adjacent exits provides an always-on second set of eyes that does not replace the sitter but covers the seconds the sitter looks away, flagging a person down or a patient at a controlled exit and routing the alert into the same workflow staff already use.
Will this let us reduce security officers or nursing staff?
No, and a hospital should not represent it that way. Emergency department staffing is constrained by labor supply, reimbursement, and the department's inability to control its own volume, not by detection capability. Regulators will not accept detection technology as a substitute for required staffing or for clinical observation. What the detection layer does is make the existing team faster and better-positioned: officers and nurses respond to a precise location with a time stamp the moment an event is detected, instead of finding it on a round or by chance. The value is in coverage and response speed, not headcount reduction.
How does a detection layer support Joint Commission and state workplace-violence requirements?
The Joint Commission's workplace-violence-prevention standards require accredited hospitals to operate a documented program with worksite analysis, incident tracking, and follow-up. A detection layer contributes a time-stamped, location-specific record of incidents and responses that strengthens the worksite analysis and incident-tracking the standard requires, and it demonstrates an engineering control rather than a policy alone. As more states enact health-care workplace-violence-prevention mandates, an always-on detection capability increasingly functions as a compliance artifact, objective evidence that the hospital implemented a real environmental control at the point of highest risk, which a passive recorder reviewed only after an incident cannot provide.
Do we have to replace our cameras or our video management system?
No. The platform layers on top of the existing IP camera network and integrates with the hospital's existing video management system. Detection runs on a dedicated rack-mounted appliance installed in the hospital's own server room. No camera replacement, recabling, or network re-architecture is required for a typical deployment, and the hospital continues to own its video and its retention policy. Initial coverage of triage, the waiting room, and the priority hallways is typically reached within days of appliance installation, followed by a per-camera tuning period to calibrate detection zones to the department's layout.
Continue the research
This playbook covers why the emergency department is its own security environment, what the violence-and-falls convergence data shows, and where AI detection earns its place inside the department. For deeper reading on the adjacent frameworks and the underlying technology:
- Healthcare Workplace Violence: The AI Detection Playbook, the hospital-wide companion to this department-level playbook.
- The Reimbursement Cliff of Inpatient Falls, the ROI economics of a fall once it happens.
- The Joint Commission 2026 Workplace Violence Standards, the accreditation requirements an emergency department program must satisfy.
- AI Fall Detection: How Computer Vision Identifies Falls in Real Time, the technical reference on the detection pipeline.
- Fall Detection solution page, the IntelliSee fall-detection modality and supported systems.
- How IntelliSee works, the platform architecture and the existing-camera retrofit model.
- Request a risk assessment to model the ED convergence framework against your own incident log and department layout.
More intelligence like this
New IntelliSee research drops monthly at most. Subscribe and get the next sector playbook, technology briefing, or threat intelligence report in your inbox the day it ships.
Request a Risk Assessment
Talk to an IntelliSee security specialist. No sales pitch — a structured conversation about your environment, your threat profile, and whether computer vision is the right fit.
Request a Risk Assessment