Walk the exhibit floor at GSX this year and you will undoubtedly encounter one term repeatedly: artificial intelligence. AI-enabled video analytics. AI-powered access control. AI-assisted monitoring. Generative AI. Predictive intelligence. Autonomous systems. The list will be long.

That emphasis is understandable. GSX 2026 itself reflects how important AI has become to the security industry. The conference includes more than 500 exhibitors and 200 educational sessions, with X Stage programming featuring AI and emerging technologies, a hands-on LLM security workshop, and a keynote from former Google Chief Decision Scientist Cassie Kozyrkov focused on applying AI and data-driven decision-making to security challenges.

AI will transform physical security — that much is certain. The harder question is determining which products represent meaningful advances and which simply attach "AI" to capabilities that have existed for years.

As you walk the GSX exhibit floor, here are seven questions worth asking before getting too excited about any "AI-powered" security product.

1. What exactly does the AI do?

This sounds obvious, but start here. Ask the vendor to describe specifically where AI is being used and what it accomplishes that conventional software cannot. Does the system recognize objects? Identify anomalous behavior? Correlate information from multiple systems? Search video using natural language? Prioritize alarms? Generate incident reports? Recommend actions to an operator? "AI-powered" describes a technology, not a capability. If the vendor cannot clearly explain what the AI does, how it works within the security workflow, and why AI improves the outcome, you may be looking at marketing rather than innovation.

2. What problem does it actually solve?

Technology demonstrations tend to emphasize capabilities. Security practitioners should focus on problems. Ask what operational problem the product eliminates or materially improves. Does it reduce nuisance alarms? Decrease investigation time? Improve detection probability? Reduce operator workload? Accelerate response? Allow fewer people to monitor more assets? Better yet, ask the vendor to quantify the improvement. A compelling demonstration is interesting. A measurable operational outcome is valuable. The objective is to improve security performance, efficiency, resilience, or some combination of the three — deploying AI is a means to that end, never the goal itself.

3. How accurate is it and under what conditions?

AI demonstrations usually occur under favorable conditions. Real security environments rarely cooperate. Ask about false-positive and false-negative rates and, importantly, how those rates were established. Then ask what happens when conditions change. How does the system perform at night? In rain or snow? In crowded environments? With partially obstructed subjects? At different camera angles? Across different demographics, facilities, or operational environments? A system that performs extremely well in a controlled demonstration may perform very differently when deployed across hundreds of cameras at a complex facility. The relevant question is: "How accurate will it be in my environment, and what are the design requirements to optimize performance?"

4. What data does the AI need and where does that data go?

AI systems are fundamentally data systems. Understanding what information a product collects, processes, retains, and shares should therefore be part of the security design process. Does processing occur at the edge, on premises, or in the cloud? Is customer data used to train models? How long is information retained? Can data leave the customer's environment? Who owns generated metadata? What happens to that information when the contract ends? These are physical security questions as much as IT questions. As physical security platforms become increasingly cloud-connected and AI-dependent, data governance is becoming part of physical security architecture.

5. What happens when the AI is wrong?

This may be the most important question on the list. Every AI system will eventually make a mistake. The critical issue is what happens next. Does the system flag something for human review? Does it recommend an action? Can it automatically initiate a security response? Can an operator understand why the system reached its conclusion and override it? The consequences of an AI incorrectly categorizing a video clip are very different from those of an AI autonomously denying access, dispatching security personnel, or escalating an incident. As AI becomes more capable, security organizations will need clearly defined boundaries between AI-assisted decisions and AI-authorized actions.

6. How does it integrate with the rest of the security ecosystem?

A remarkable AI capability operating in isolation may have limited operational value. Ask whether the product integrates with your video management system, access control, intrusion detection, visitor management, identity platforms, incident management, threat intelligence, or other enterprise systems. Then dig deeper. Are those integrations based on open APIs or proprietary interfaces? Is data exchanged bidirectionally? Can the AI initiate workflows across other systems? Some of the most valuable AI applications will likely emerge from correlating information that historically existed in separate systems — making integration potentially more important than the intelligence of any individual device.

7. Can you prove the business case?

Finally, move beyond the demonstration and discuss economics. What does the technology cost to deploy, integrate, license, maintain, and operate? Then compare those costs with measurable benefits. If an AI platform reduces alarm volume by 80 percent, cuts investigations from 30 minutes to five, or allows one operator to effectively monitor what previously required three, the business case may be compelling. AI that produces an interesting new capability without improving security outcomes or operational efficiency may simply create another technology platform to manage. That distinction will become increasingly important as organizations accumulate AI-enabled products.

Look Beyond the Label

GSX has always provided an opportunity to see where security technology is heading. With AI becoming increasingly prominent throughout the industry and throughout GSX programming itself, this year's exhibit floor should provide an especially useful glimpse into the next generation of security operations. Some of what we see will likely represent significant advances. Some will be incremental improvements. And inevitably, some conventional technologies will have acquired an AI label since the last trade show. The challenge for security professionals is distinguishing among them.

So as you walk the GSX floor, ask vendors what security decision their AI makes better, what operational problem it eliminates, and what measurable outcome it improves. Those answers will tell you far more about the future of a product than the letters "AI" printed on the booth.