AI in Security Technology:
Application with Purpose
The future of security is not autonomous technology alone, but intelligent systems guided by experienced human judgment.
AI in Security Technology:
Application with Purpose
AI has become the touchstone of today’s tech world and the fascination of popular media.
The trend is that we must build more by any means possible as soon as possible. As companies rush to capitalize on seemingly unlimited opportunities, the question arises, are we pursuing innovation for innovation’s sake or is there application with purpose?
For reference, my experience is in the physical security space, having founded a security technology company over 20 years ago. While we have innovated and adopted many security technologies over those years, the push for AI is unprecedented in our space.
Candidly, when I first began researching AI in our space I was skeptical. The early iterations did not impress and it felt threatening to our status quo. With time and additional research I have changed my thinking. Today AI offers efficiencies when properly implemented and coupled with human decision making delivering quality outcomes. The challenge is and will continue to be how we choose to use these new tools.
In the rush to market, a plethora of AI security solutions, most offered by start-up companies backed by venture capital have appeared. In 2025 we evaluated seven of these offerings and discovered that the providers confidence levels in their AI generated outcomes did not meet our standards. We found stated confidence levels from 90 to 95 percent, while we are accustomed to 99%+ confidence (read success) in our own outcomes.
Why does this matter? Security decisions carry asymmetric risks, meaning that the consequences of incorrect decisions are not evenly balanced. Our clients expect us to detect and deter criminal behavior at a very high success rate bordering on perfection. 90 to 95 percent confidence equates to 5 to 10 percent error. Potential losses exist in that 5 to 10% error rate. It is imperative that humans close this gap. I cannot speak for other industries, whose tolerance for exceptions may be much lower, but in the physical security space, even 1% error rates are the cause for much consternation.
We noticed that the offerings that we considered promised “much lower or no human interaction”. The proposition for a company like ours is that by using AI we can dramatically reduce staffing levels in our SOC, presumably increasing our company’s margins and/or improving our pricing models. While that is certainly enticing, our margins are, today, within our expectations and above all else we value our client relationships. This creates a “risk/reward” conundrum that no doubt will become more compelling over time.
What is missing, in my view, is context. A “person present” alert is certainly useful, but it does not tell the full story. Is the person an intruder or a client representative on site? The reality is that most, if not all of the sites that we secure do have expected client activity that must be considered. Our challenge is that we want to deter bad actors without disrupting our clients’ workflow.
In response, we have chosen to develop our own AI solutions internally by targeting specific logical problems. Our goals are to create solutions that 1.) meet our confidence expectations, 2.) provide efficiency, and 3.) enhance our team’s ability to make correct decisions. The following table lists examples of our implementation of AI to this point.
Our “line in the sand” is that critical decisions must still be made by our trusted SOC Teams.
There are additional aspects of any successful security solution in addition to AI technology and they are foundational. We have learned that while tech (to include AI) is indeed a cornerstone of the work that we do, it is not, in and of itself, a complete solution. There are four equally important cornerstones that must be in place for long-term success.
Systems Engineering or the end to end application of tech & associated systems. This includes thoughtful design, systems integration, and validation of an overall solution. You cannot expect tech to work well if the associated systems that it interacts with are flawed.
Systems Care: The truth is that all systems, from cameras to connectivity to intelligence, fail over time. It is critical that component viability is assured in near real time. Further, when a component has failed, it must be recognized and addressed in an expedient manner. This requires systems status monitoring coupled with real-time response. No one wants to learn that their VMS has failed when trying to retrieve video.
Proper SOC Engagement: We have learned that the thoughtful use of AI can indeed provide efficiency. However, the gravity of our decisions require that a qualified human must make the final call. It is one thing to minimize false alerts but quite another to decide “friend from foe”. Today, we feel strongly that AI generated outcomes should be reviewed by human agents for final action. I expect that fully autonomous AI solutions will advance over time but today they lack the acuity of a human.
Actionable Intelligence: Our solutions generate significant data that has proven useful to our clients. Today we provide meaningful security insights via a broad group of reporting methods that our clients use to manage their security landscape. These include predictive analysis on a local and enterprise levels.
In summary, we use AI to solve logical problems when our confidence in the solution is at a very high level. These tools allow us to perform baseline tasks, such as alert processing, systems care, and site condition validation more efficiently while supporting faster and more informed decision-making.
Looking forward, AI will continue to evolve and we are excited to influence that. In today’s high-consequence environments, accountability still belongs to people. The future of security is not autonomous technology alone, but intelligent systems guided by experienced human judgment.
AI cannot, today, replace the acuity of the human mind. The weight of our decisions dictate that, in the end, we must make those decisions.
In physical security, confidence levels are not theoretical.
In 2025, RSS evaluated seven AI security offerings and found that provider confidence levels in AI-generated outcomes did not meet RSS standards. Stated confidence levels ranged from 90 to 95 percent, while RSS is accustomed to 99%+ confidence in its own outcomes.
AI Generated Security Solutions
Agentic / machine learning assessment
Lower cost potential
Lack of context in decision making
Potential incorrect assessment of critical events
Traditional Security Solutions
Analytic alerts with human assessment
Highly accurate outcomes
Higher cost potential
Potential shift in market pricing
AI should enhance human decision-making, not replace it.
Meet Confidence Standards
AI implementations must support the exceptionally high confidence levels expected in physical security environments where mistakes carry real-world consequences.
Create Efficiency
Automate repetitive and time-consuming tasks, allowing SOC personnel to focus their attention on activities that require judgment and experience.
Improve Decision Quality
Provide better information faster so trained professionals can make more informed decisions with greater situational awareness.
Security technology should deliver confidence, not uncertainty.
RSS combines AI, advanced analytics, and experienced human judgment to create security programs built for real-world outcomes. Discover how a layered approach can improve security performance while reducing operational risk.