AI is advancing rapidly, creating both opportunities and challenges. How can Physical Security leverage AI effectively?
Q: Our senior executives are highly interested in AI and want to know how Physical Security will utilize it. How do I address AI’s fast evolution and presence across many products?
A: This is a core challenge for facility security professionals. Although AI continues to evolve, traditional risk-based approaches to selecting and deploying security technology remain relevant. They need to expand to account for AI’s growing capabilities.
Preparing for Productive AI Use
AI is a broad topic and requires a longer than usual column to cover the important aspects of effectively utilizing emerging AI-enabled physical security devices and system capabilities.
Effectively leveraging emerging AI requires understanding five key elements:
- AI as a Technology Category. AI is not a single product; it includes software, cloud platforms, and embedded chipsets. Many security devices now feature built-in AI for tasks like machine learning (ML) and deep learning (DL), raising questions about risks and errors. Security practitioners must consider these to ensure AI-enabled tools are safely deployed. Video perception analytics now include generative AI capabilities (ChatGPT on a smaller scale), which may require specific organizational approval.
- Closing Longstanding Security Gaps: Breakthrough products that integrate seamlessly with existing systems can now address Decades-old security gaps. AI enhancements can effectively close these gaps, mitigating vulnerabilities that could not be addressed cost-effectively.
- Infrastructure Transformation: Traditional “rip and replace” methods are expensive and disruptive. New AI-compatible systems support gradual upgrades, allowing phased improvements without costly overhauls.
- Aligning with Business Goals: Emerging AI technologies should be evaluated based on their potential to reduce security risks at a manageable cost while supporting security, safety and broader business operations. New security system data insights can benefit security and business functions, making a case for aligning physical security technology with business objectives.
- Data over Devices: AI-enhanced security systems now generate data that can improve business processes, mandating a shift from focusing on individual security devices to data-driven security and business operational insights.
A Watershed Moment in Physical Security
The physical security industry is pivotal, marked by a shift from rules-based limited automation to real-time, intelligent automation that provides instantly actionable data and ongoing real-time situational awareness.
Decades-Old Security Gaps
For 50 years, physical security technology progressed very gradually. Today, advancements that used to take a decade now occur in just a year, requiring a fundamental shift in thinking. Past systems lacked intelligence—humans analyzed sensor data or monitored scenes directly. We named systems based on how humans used them rather than the systems' functionality.
Misnaming Security Technology
Thus, historically, technology names have been misleading. For example, intrusion detection systems didn’t determine if an intruder was present. They sensed changes in sounds and/or temperatures but required humans to determine if an actual intruder was present.
Access control systems should have been called “electronic door lock systems.” One early proximity card access control system company named itself Sielox, which stands for Secure Industrial Electronic Locks.
Facility security personnel still must follow up on access control system door alarms like “forced open” and “open too long”, the majority of which are not security incidents. Closing all access control gaps would require monitoring all access events, including “access granted,” to account for tailgating (one individual sneaking in behind another) or piggybacking (an authorized user holding the door open for an unauthorized user.)
For most organizations, tailgating/piggybacking detection systems have been too expensive to cover all portals that warranted them. Furthermore, they still require human video analysis to determine whether an officer should be dispatched to a tailgating location because the person attempting unauthorized access may continue to pose a threat.
AI has changed that. Great examples are the Rock (indoor) and RockX (outdoor with intercom capability) devices from Alcatraz AI, which perform facial authentication rather than facial recognition. They don’t store facial images but build a facial mathematical model that they compare with other models in their database. A facial image cannot be reconstructed from the mathematical model, and therefore, the products can be used where facial recognition has been prohibited. If tailgating or piggybacking is attempted, the devices can instantly send an image of the unauthenticated individual to a responder for follow-up, eliminating the need to examine recorded video, which can delay the response.
This is an excellent example of using Machine Learning at the device level to close a 50-year-old functionality gap in existing access security system deployments. Furthermore, recent AI models can be applied across multiple devices and sensors at the system level to provide deep real-time situational awareness that enables highly informed instant response.
Game-Changer AI Models
Pairing newly emerged Large Language Models (LLMs) and Large Multi-modal Models (LMMs) is revolutionizing security operations by enabling real-time, cross-media analysis of text, images, and audio. These AI models deliver plain-language, actionable descriptions of unfolding situations, eliminating delays from human review of alarms and recorded video before response actions can begin. For example, they can detect unauthorized access attempts at multiple secure entry points, enhancing situational awareness across a site.
An alert might read: “An unauthorized individual in dark blue pants and a light blue shirt is attempting access to the R&D lab at multiple doors.” This analysis draws on access control and video surveillance data. AI could extend the scenario by adding: “The access card being used isn’t registered in the system. The individual drove onto the property 15 minutes ago in a red pickup truck parked in row 5. The person attempted to tailgate into R&D behind Gretchen Smith, who denied entry.”
In this case, a security system rule could dispatch one officer to the latest entry point and another to the trespasser’s parked vehicle, equipping a detailed activity description and the clearest image of the unauthorized individual.
Real-Time Situational Awareness
This level of situational awareness is transformative. Previously, security staff could take up to 20 minutes to correlate alarms with video, often missing real-time context. Now, AI-enabled systems provide instant, actionable updates as incidents unfold. Rather than replacing personnel, this AI functions as a force multiplier for existing personnel, potentially doubling or tripling a small team’s response capacity across simultaneous incidents or multiple threat actors. For the first time, security teams can rely on real-time, detailed situational insights, regardless of the scale of activity.
Adding AI and Automation Capabilities to Legacy Systems
These capabilities aren’t futuristic; they can be added to existing security systems, a key value of Soloinsight, a leader in security automation and access control solutions. Ambient AI and Camio are two companies whose customers already have AI capabilities like those described in this article.
James Connor, Ambient’s Head of Corporate Engagements, noted, “The job of a security officer is to ‘observe and report.’ LLMs and LMMs make this job faster and more effective by linking sensor data, including video, in real-time, enabling officers to act quickly rather than retrospectively.”
Carter Maslan, CEO of Camio, emphasized, “For the first time, we can address risks with the equivalent of 1,000 experts fully trained in our policies from day one. The shift from simple classifier analytics to LMMs capable of understanding video, images, and data, as well as security policies and procedures, allows us to tackle risks that previously required a human to prioritize and resolve. These AI capabilities emerged only six months ago, driven by billions in investments outside the physical security industry. With them comes a mandate to harness these rapid AI advances for security and additional business operations capabilities.”
As one Camio customer said, “Every camera has its own post orders!”
Connor often states, “Modern innovations should not just bring forth newer capabilities but also harmonize with existing legacy systems, prioritizing scalability and upholding standards for security and privacy.” President of Soloinsight, Farhan Masood adds, “The essence of modern security lies in their adaptability. Legacy systems reach their potential when integrated with advanced automation and identity management solutions.”
Developing AI-Enabled Security Operations Capabilities
Facility security practitioners can now reassess their needs by evaluating past and potential risks through an AI-enabled perspective. For instance, consider, “If I could post a security officer 24/7 at every key vantage point, what should they monitor, and what data would help inform their actions?” This approach involves evaluating camera coverage and determining which systems contain relevant information.
In serious incidents, like potential workplace violence, AI can assist by identifying individuals with restraining orders who should be barred from entry, enhancing at-risk employee protection. Routine nuisances, such as unauthorized parking in reserved spots, can also be prioritized by risk and frequency. With AI, such scenarios can be automatically evaluated against the relevant policies and procedures, allowing faster, data-informed, appropriate responses.
For example, a VIP visitor arriving late for a high-level executive meeting could be allowed to park in the reserved space for the duration of the meeting, instead of broadcasting the parking violation over the public address system, which would interrupt the meeting and embarrass the VIP as well as the meeting host. This is an example of security using AI to intelligently apply policies and procedures harmoniously to the business, which requires a more comprehensive evaluation of security operations and situation response that has been needed in the earlier security technology era.
Data-Driven Security and Business Operations
As businesses become more data-reliant, AI-enhanced security systems provide data insights beyond security, boosting efficiency across functions like staffing, retail responsiveness, marketing campaign evaluation, and manufacturing area safety compliance monitoring.
By comparing current physical security capabilities with new AI potentials, security can create a risk-prioritized, cost-effective strategy that maximizes existing personnel resources and adds value across business operations.
What once was impossible is now within easy reach with AI.
