vietnamese crab exporter
Commercial Content

Exploring how AI and human judgment could work together in security screening

Karen Koehler

AI’s role in physical security is becoming a practical question of how teams detect potential threats and respond. A 2025 ASIS International report found that 23% of surveyed security professionals said their organizations used AI or machine learning in surveillance for object detection or motion tracking, while 19% reported using it to flag unusual situations. These applications bring human oversight into focus. When technology identifies a potential concern, who reviews the alert and decides what happens next?

That responsibility extends beyond reviewing individual alerts. Institutions also need to establish how AI fits into their security procedures and who remains accountable for decisions. The National Institute of Standards and Technology is examining these broader concerns through its work on trustworthy AI in critical infrastructure, including a concept note released in April 2026.

Its existing AI Risk Management Framework calls for clearly defined responsibilities, documented human oversight, and testing before deployment and throughout operations. For entry screening, these principles suggest that evaluating technology should include how well staff understand its alerts and how clearly procedures guide their response.

NIST also recommends assessing performance under conditions similar to those in which a system will operate. At a busy checkpoint, that means looking beyond how quickly people pass through. A fuller evaluation would consider how alerts are reviewed, how staff manage follow-up screening, and whether the process supports informed decisions during periods of heavy traffic.

For Chris Ciabarra, AI’s role in security begins with the people operating it. He is the co-founder and chief technology officer of Athena Security, a company that develops AI-based screening technology. He believes its value lies in helping security personnel notice potential concerns while retaining responsibility for consequential decisions. “Anything dealing with security should have human oversight,” he says. “It should never just act on its own.”

For Ciabarra, a busy school entrance illustrates the challenge. Officers may be monitoring arrivals, reviewing alerts, and ensuring that screening procedures are followed. He sees potential for AI to direct attention toward something requiring review, such as a possible object in an X-ray image or someone bypassing a screening lane. In his view, these prompts should support an officer’s assessment.

His preferred design principle is to give each AI component a clearly defined task. One might assist with image review, while another could flag a missed procedural step. “The important thing is you limit what they can do,” Ciabarra explains. “You don’t want one agent doing everything.” He explains that narrower responsibilities could make systems easier to supervise.

That preference should still be paired with verification. The same report by NIST recommends documenting an AI system’s limits and evaluating its performance under relevant conditions. A focused task therefore remains something to test, rather than evidence of reliability by itself. Ciabarra similarly emphasizes checking whether equipment works properly as part of the checkpoint’s routine, alongside reviewing how personnel carry out screening.

He also anticipates more centralized oversight, with alerts from several checkpoints reaching human reviewers in one location. Ciabarra says his company already supports centralized alert review for its X-ray baggage screening platform. His broader vision extends that arrangement to additional checkpoint functions, although he presents this as a future direction rather than an established model for every facility.

In that proposed arrangement, he envisions reviewers assessing alerts and directing the next screening step. His emphasis remains on preserving human involvement as automation expands. For schools considering such an approach, the NIST framework’s focus on defined responsibilities offers a relevant consideration. Institutions would need to establish who reviews information, who responds locally, and how those duties fit together.

Ciabarra’s perspective suggests two possible directions. With clear boundaries and trained reviewers, AI could help staff focus attention where it is needed. If institutions treat automated output as sufficient on its own, they could weaken the human oversight he considers essential. The difference, in his view, rests partly on how technology is incorporated into everyday procedures.

For leaders evaluating these tools, his message points toward a practical starting point. Define the task, establish who reviews the result, and determine how performance will be checked. Ciabarra sees the purpose of adoption as strengthening the team’s ability to act. Keeping that purpose visible could help institutions assess new capabilities without allowing responsibility to become unclear.

To learn more about Athena Security’s approach to AI-assisted screening, visit Athena Security.