Agentic Physical Security for Hospitals: Threat Detection from the Emergency Entrance to the Locked Unit

See what an operator sees when a threat signature recognizes a behavior
TL;DR
A hospital's cameras record through the night and its doors log badge reads, yet after visiting hours the emergency entrance still opens to anyone who needs it. The moments that matter are easy to miss: two visitors coming to blows in the entrance vestibule, an employee down in an empty service corridor, a visitor lingering at a locked unit door. Threat detection turns each one into an alert that a small security team can confirm and act on while the situation is still forming.
Picture the emergency entrance of a community hospital at two in the morning. The main lobby locked its doors after visiting hours. Here, though, the sliding doors open for anyone who walks up to them, the cameras over the ambulance bay are recording, and no alarm has reached the security control room since midnight. Nothing is broken, and this is the one door in the building that is never meant to lock. The building is doing what a hospital does, which is staying open to whoever needs care at that hour.
That openness is the point of the building and the hardest part of hospital physical security. A recording camera and an open door cannot say which situation inside is escalating, who has gone down where no one is looking, or who is waiting at a locked door for someone to open it. Turning people away is not an option. The answer has to come from noticing the few moments that matter inside a building that stays open, early enough for a person to act on them.
What does hospital physical security need to detect in an open building?
Hospital physical security looks after patients, visitors and staff, along with the medications and equipment a hospital depends on, in a building that never closes. Much guidance on healthcare physical security centers on checklists and layered controls, from visitor management to locked units. That advice is sound for deciding who gets in. What no lock or checklist reaches is the moment after people are inside.
At two in the morning the officers on shift may be on a patient watch in the emergency department, on an escort to a car, or on a round of the parking structure. None of those posts comes with a view of the camera feeds, and a fall in an empty corridor makes no sound that reaches anyone. Adding cameras gives that team more footage to review the next day, not more attention in the moment. For a small healthcare security team the gap is attention, not coverage. The useful job for AI is to reason over the video the cameras already capture and put the few moments that matter in front of a person.
Ambient.ai is an Agentic Physical Security platform that works with a hospital's existing cameras and access control. It reasons over live video through threat signatures, each pre-trained to recognize a specific behavior. They range from early indicators, such as a person loitering outside a secure door, to incidents in progress such as people fighting. When a threat signature recognizes its behavior, an operator receives an alert with the video moment and the location. Where access control is connected, Ambient.ai also correlates that video with badge events at the same door. The mechanics are covered in more depth in how Ambient.ai's threat signatures read behavior. With each alert, the security team receives the information it needs for situational awareness, and Ambient.ai helps automate the response the team has defined.
What follows are three moments in that building, from the emergency entrance through the corridors behind the wards to the door of a locked unit.
At the emergency entrance after midnight, a dispute between two visitors is the first moment to notice
Consider two visitors in the vestibule outside the emergency entrance, a little after two in the morning, both waiting for news about the same patient. Their voices rise, and anyone who has worked in emergency department security knows how often a dispute like this settles by itself, because people reach that door frightened and exhausted. Sometimes it does not, and a minute later the two are trading blows in the vestibule. The officer posted inside may be in a treatment room on that patient watch, with the vestibule out of sight behind two sets of sliding doors.
Ambient.ai has a threat signature built for this moment: People Fighting, which recognizes a physical fight between people from video alone. The alert reaches an operator with the threat signature name, its severity rating and the video moment, so the operator can see at a glance whether this is a confrontation that needs an officer. The officer nearest the entrance can then reach the vestibule while it is still contained to the two people involved, rather than after it has spilled through the doors. A fight there can also draw in a registration clerk or a triage nurse who steps out to help. That makes an officer's early arrival part of hospital workplace violence prevention, not only a matter between two visitors.
In the corridors behind the wards, no one is positioned to notice a person who has gone down
Behind the wards runs a second hospital, made of service corridors, loading areas and the long passage that connects the kitchen to the dock. Imagine a night-shift employee walking that passage alone at four in the morning carrying a box of supplies, with nobody scheduled to come through for an hour. If they slip or collapse, the corridor tells no one. The first person to find them may be the morning delivery driver.
The threat signature for that corridor, Person Fallen Down, notices a person on the floor from video alone and puts the alert in front of an operator. The operator looks at the moment and judges it, and may find someone working at floor level beside a cart, which settles the question quickly. If the person is down and not getting up, the nearest officer can go straight to that stretch of corridor instead of finding them on the next round.
It's important to note that Ambient.ai approaches detection with privacy by design, and no facial recognition or PII is used for detection, which matters in a building where guidance on cameras and patient privacy already shapes how video is handled.
At the door of a locked unit, the question is who is lingering outside and who just followed someone in
The door of a behavioral health unit on the evening shift is one of the most carefully held doors in the building, and the moments that matter there are small. A visitor has been standing near it for some time without knocking or using the intercom. For that moment there is a threat signature, Person Loitering Outside Secure Door, which notices the waiting itself from video alone. Staff at the unit desk might notice that on a quiet evening and miss it during medication rounds, and the threat signature raises it as an early indicator either way.
A little later a clinician badges through the door, and the visitor steps in behind them before it closes. If the door is integrated with access control, the Tailgating threat signature pairs the clinician's badge event with video of a second person entering on it. At a unit door, the access control a hospital relies on is meant to let clinicians move quickly, which is exactly what makes a follower through a controlled door easy to miss.
Because the alert reaches an operator while the situation is still forming, a call to the unit lets a nurse at the desk step over and ask which patient the visitor came to see. The answer may be a relative who arrived late and did not know the unit buzzes visitors in. Sometimes it is a visitor the care team has asked to stay away, or one carrying something staff would have checked at the desk. A question asked near the door is a far easier conversation than the same one asked at a patient's bedside.
An alert earns its place when the officer closest to the door can act on it
In a health system, these alerts may all reach one operator in a control room that also answers for the outpatient clinics, medical office buildings and parking structures. Each alert carries the name of the threat signature, a severity rating of low, medium or high, an AI-generated description of the scene, the video moment and the location. It arrives with floor plan context that shows where it happened, and the operator reviews the video moment instead of scrubbing footage. The scene description travels with the alert, so the operator can brief whoever goes to the door, whether an officer or a charge nurse on the unit, on what to expect.
At the emergency entrance, an audio message through speakers connected to Ambient.ai can direct behavior on site while the officer is on the way. For the fall in the service corridor, targeted dispatch mobilizes the right on-site officer, and the control room follows the status updates on the incident timeline as the officer responds. At the unit door, recommended actions arrive as options for the operator, who can escalate in one click or let a workflow the team defined in advance notify the people the team named. That is how Ambient.ai supports the response while every decision stays with the team.
Once an incident is over, the same video supports the investigation. Operators can search recorded video in plain language, see results for the same person or vehicle grouped together, and gather clips from across cameras into one case with an AI-generated narrative. Because detection, response and investigation share one console across every connected site, Ambient.ai works as a fully integrated platform, and an alert from an outpatient clinic across town is handled like one from the main entrance.
What should threat detection do for a small security team covering an open building?
Each of the four asks below works on a first vendor call or in an RFP, and none of them depends on which platform is answering.
- It reads the cameras and doors the building already has. Ask for the camera and access control models it supports in writing, and check them against what is installed across your sites.
- Any officer can read an alert without a phone call. Ask to see a sample alert with the detected behavior, the severity rating, an AI-generated scene description, the location and the video moment, and ask whether that description can be shared with whoever responds, clinical staff included.
- Detection reaches the places a hospital cannot lock. Ask which detections cover the emergency entrance, the back-of-house corridors and the door of a locked unit, not only the perimeter.
- Response follows your own protocols, including who on the clinical side is told. Ask whether your procedures can run as workflows that escalate and dispatch automatically, and expect your team to define and control every one.
Hospital threat detection that passes all four gives a small team a way to notice what matters inside an open building, early enough for a person to act.
How Ambient.ai surfaces the moments between the entrance and the locked unit
Every threat signature in these scenes comes from Ambient Threat Detection, one of the components of the Ambient.ai platform. People Fighting, Person Fallen Down and Person Loitering Outside Secure Door work from video alone. For doors integrated with access control, Ambient Threat Detection correlates live video with badge events, so the clinician's badge and the visitor who followed it arrive as one Tailgating alert. Throughout, the security team decides the response.
Tonight, in your own building, how soon would anyone know that two visitors had started trading blows at the emergency entrance, that a night-shift employee had gone down in a service corridor, or that the visitor waiting outside a locked unit had just followed a clinician inside?
Key Takeaways
A hospital's emergency entrance stays open by design, so much of the risk sits in what happens inside, not only in who enters.
Threat detection reads a fight, a person down and loitering at a secure door from video alone, and tailgating at doors linked to access control.
Each alert carries the behavior, severity rating, scene description, video moment and location, the team stays in command of the response, and the same video supports the investigation afterwards.
Good detection works with installed equipment, explains every alert, reaches the spaces a hospital cannot lock and runs on the team's own protocols.
.webp)