Identifying the Threat, Not the Patient: A New Approach to Physical Security in Healthcare

The following is a guest article by Shikhar Shrestha, Co-Founder and CEO at Ambient.ai

In 2026, healthcare organizations face a difficult balancing act in how they secure their facilities. Providers must keep hospitals welcoming and dignified for patients seeking care, while simultaneously implementing the rigorous protocols needed to protect staff from a rising tide of workplace violence. From the high-pressure environment of the emergency department to pharmaceutical storage areas that often house millions in controlled substances, hospitals are full of at-risk zones. And healthcare workers are now five times more likely to experience workplace violence than those in any other sector, with incidents costing the industry billions annually. Yet traditional security, built on passive cameras, static rules, and human operators stretched thin, is often at odds with the fundamental right to patient privacy.

A new approach is emerging to resolve this tension: Agentic Physical Security, a model powered by AI that reasons over time and context. Unlike the video analytics of the past decade, which rely on single-frame object detection or cloud-based pattern matching, agentic systems do not simply recognize objects; they understand behavior. They do not just alert; they assess and act. And critically for healthcare, they can do all of this while focusing on what a person is doing rather than who that person is.

The regulatory landscape, defined by HIPAA compliance and the Joint Commission’s 2026 National Performance Goals, requires hospitals to move beyond passive surveillance toward active workplace violence prevention. Meeting those standards without compromising the dignity of patients or the trust of the community demands a shift in how the industry thinks about security: identifying dangerous actions while intentionally ignoring the identity of the person performing them. That is the premise of Agentic Physical Security, and it is the only sustainable path forward.

Why Human Vigilance is Not Enough

The modern medical center is one of the most difficult environments in the world to secure. Unlike a controlled corporate office, a hospital is a high-traffic, 24/7 public utility. Emergency departments must remain accessible to all, pharmaceutical areas require strict access control, and staff navigate emotionally charged interactions every hour of every day. This openness is essential for care, but it creates a monitoring burden that cannot be solved by human eyes alone.

It’s not a people problem. It’s a systems problem. Research from the IAHSS Foundation confirms a critical failure point in traditional security: even the most dedicated operator experiences a sharp decline in cognitive focus when monitoring live video. The National Institute of Justice has found that humans lose roughly 95% of their attention on video monitors after 20 minutes. When scaled across a hospital campus with thousands of cameras, the probability of missing a precursor to violence, such as loitering in a restricted area, unauthorized entry, or reconnaissance behavior, becomes a statistical certainty.

Regulatory pressure has meanwhile reached a tipping point. With the Joint Commission designating workplace violence prevention as a National Performance Goal for 2026, and states like California and New York enforcing strict compliance mandates, a reactive security model is now a significant liability. It is no longer sufficient to have recorded footage after an incident. Facility managers must demonstrate that they have proactive systems in place to detect and mitigate hazards before they escalate into injury. The objective for the modern Security Operations Center (SOC) is to transform video feeds from a source of constant noise into a precise, real-time tool for early intervention.

The Agentic Framework: See, Think, Assess, Act

Agentic Physical Security is defined by a continuous loop. The system sees (accurate perception across every camera), thinks (reasoning that connects signals over time), assesses (evaluating location, behavior, and intent to determine real criticality), and acts (initiating investigation or policy-driven response). This is not automation, and it is not a rules engine. It is a model that reasons about what is happening, decides whether it matters, and initiates the right response, with humans in the loop rather than in the bottleneck.

To operationalize this approach in healthcare, security leaders should anchor on three principles.

Reasoning Through Vision-Language Models (VLMs)

Purpose-built VLMs, trained on ethically sourced enterprise video, can distinguish routine activity from a legitimate threat. A janitor propping a door to clean is not a tailgating event. A grieving family gathering in a corridor is not a loitering alert. Understanding the why behind a scene, rather than matching pixels, filters out the thousands of irrelevant events that overwhelm operators and ensures that the alerts escalated to humans genuinely warrant human judgment.

Orchestration Across Multiple Signal Sources

An agentic model does not operate on video alone. It ingests and correlates signals from a growing fleet of sources across the modern hospital campus. Physical Access Control Systems (PACS) are one example; others include perimeter drones, robotic ground patrols, IoT sensors, badge readers, and building systems. By cross-validating a Door Forced Open alarm with live video, or confirming a drone’s perimeter alert against access logs, agentic systems automatically resolve more than 90% of nuisance alarms and turn an overwhelming volume of disconnected inputs into a coherent, high-fidelity picture.

AI-Augmented Response and Decision-Making

Detection only matters if the system can act on it, but acting does not mean replacing the operator. The AI engine observes the scene, builds situational context, and surfaces recommended response plan options to the operator, so decisions that used to take minutes can be made in seconds. Once the operator approves, the model automates the mechanical parts of the response, such as dispatching personnel, initiating 911 calls, triggering live audio talk-downs, and executing standard operating procedures. The operator retains full authority; the system removes the friction between decision and action.

Privacy by Design: Non-Negotiable in Healthcare

Healthcare is the one environment where security and privacy must be inseparable. Patients arrive at their most vulnerable, and every camera in the facility is, by default, a potential privacy risk. That is why an agentic model built for healthcare must be architected around privacy from the first frame, not as a setting that can be toggled on after the fact.

In practice, this means detection built around threat signatures, such as a person falling, a door propped open in a restricted pharmacy area, or reconnaissance behavior near a nursery, rather than facial identity. It means no facial recognition, no stored PII, and raw video that never leaves the customer’s environment. It means alignment with HIPAA, GDPR, and CCPA standards as a baseline, not a marketing claim. Focusing on the what and the where instead of the who is how hospitals can achieve 24/7 coverage of sensitive areas while preserving the dignity that makes care possible in the first place.

Metrics That Matter: Measuring a Proactive Posture

Beyond the architecture, it is vital to establish metrics that quantify improvement. Traditional security KPIs still have their place, but Mean Time to Acknowledge (MTTA) is the most meaningful measure of a truly proactive posture. Automating event monitoring and verification compresses the window between detection and response from minutes to seconds.

Alongside MTTA, false positive rates matter, not just as a noise metric, but as a measure of whether operators can trust what reaches them. Combined with the ability to identify precursor behaviors such as loitering, reconnaissance, and unauthorized access attempts, agentic systems give security teams the chance to intervene before a situation escalates, rather than investigate after it does.

This is the promise of Agentic Physical Security in healthcare: a model that sees everything, respects every patient, and acts only on what truly matters. The technology is here. The regulatory environment is demanding it. The question is no longer whether healthcare organizations should make the shift, but how quickly they can.