AI is making waves; it’s tantalising, exciting, and disconcerting all at once. In physical security, it promises new levels of service, but raises real questions about privacy, data and control. 

The upside of AI in physical security is huge, but only if done right. The headline message? Tread carefully, but keep moving forward. If you don’t, you can be sure your competitors will. 

Having spent many years at The Keyholding Company (TKC) thinking of new ways to transform how we operate, here’s what I think successful AI adoption really looks like for our industry. 

Pressure is mounting to adopt AI in physical security

The industry is notoriously slow-moving when it comes to tech innovation. Many organisations still rely on manual, fragmented processes, where email chains and spreadsheets are still far too common. 

The result is poor visibility, inconsistent reporting, and friction across service lines – especially in areas like guarding, where human error and low efficiency remain a challenge. 

It can’t – and won’t – go on like this. Clients now expect smarter, more transparent services, and larger buyers are demanding it. Traditional providers are under real pressure to modernise – and fast. 

From day one of my time at TKC, I’ve loved our commitment to use technology to push the boundaries of what’s possible in physical security. But AI isn’t a bolt-on; it requires a fundamental rethink of how you operate. Like any transformation, the risks are as real as the opportunities. 

As TKC’s Chief Transformation Officer, my role is to help us navigate that shift. Here’s what I’ve learned: 

AI should be applied where it actually solves problems in security operations

Many companies want to adopt AI but face daunting questions like ‘where do I begin?’, ‘how should we implement it?’ or ‘what’s going to have the biggest impact?’  

The biggest misconception about AI in physical security is that it replaces humans. In reality, the smartest applications enhance human performance, not eliminate it. The right approach is to find repeatable, high-volume, minimal subjectivity tasks and let AI do the heavy lifting.  

So, what does that look like in practice? 

Example 1: Photo verification on job reports

At TKC, we carry out 1.2 million mobile security jobs yearly. On every job, security officers take multiple photos on our mobile app. These images need to be validated for quality, location, context, and content – was the uniform correct, was the photo taken on-site, was the right room captured etc.  

Where most security firms run manual checks hours later, TKC uses AI Vision to detect errors in real time, throughout each shift.  

This dramatically speeds up quality assurance, removes manual workload, and increases the accuracy of our job reports. Soon, real-time alerts will let officers fix issues mid-job – a leap forward in proactive compliance. 

Example 2: CCTV triage

Traditionally, Control Room operators manually triage thousands of motion-triggered CCTV alerts – most of them false positives (e.g. pets, lights, shadows). 

With AI embedded into Immix, TKC’s monitoring platform, we automatically suppress alerts that meet certain parameters – detecting actual humans in the footage while filtering anything that could trigger a no-escalation ‘motion event’. This can extend to automated health checks, picking up cobwebs on the camera lens, or flagging unauthorised changes to camera direction. 

For one client, AI reduced 6,500 alerts to just 500 requiring human review – a 90% efficiency gain.

Chris Russell

The goal isn’t just to ‘add AI’, but to ensure human operators only deal with genuine threats, reducing fatigue, improving service, and delivering real value from surveillance systems. 

AI adoption requires significant system changes, but don’t feel rushed and dive in too deep

Many security firms will struggle, by trying to bolt AI onto legacy systems without first digitising their operations. At TKC, we’ve taken the opposite approach – building a foundation that allows AI to integrate seamlessly as the tech evolves.  

This means we can plug-in AI services without fundamentally changing our underlying tech architecture, letting us build fast, iteratively, and with less senior resource. We’re building hybrid systems that combine the strengths of both human and digital labour, each acting as both an enhancement and a failsafe to the other.  

Full AI adoption requires major system changes, but it’s not about diving in the deep end. Security teams need to work hard on assessing where AI adds value to their operations and must resist the urge to ‘do AI’ for its own sake. It’s about taking baby steps and building up, learning what works, and measuring impact.  

AI moves fast – but rushing into half-baked solutions will only pull you back when you should be moving forward. Take the time to get the foundations right. 

Managing the risks is key 

An AI-powered future is exciting, but it raises valid questions: Can we trust AI not to make mistakes? What happens to the vast data we feed it? Will moving away from ‘good enough’ legacy systems backfire? 

These fundamental concerns must be addressed if the industry is to progress with AI, because stakeholders and decision-makers will always weigh risks against benefits. 

That’s why our AI adoption pathway is risk first, benefit second, to build that essential foundation of confidence before scaling. Here are my top three focus areas: 

1. Data security & privacy 

TKC use the OpenAI Enterprise API for our AI integrations. It’s backed by robust data security controls such as high-level data encryption and secure access controls, and regular InfoSec audits (SOC-2). 

None of the data we handle is used to train models – a common concern. Instead, we benefit from models already trained on trillions of data points, letting us benefit from datasets without having to provide inputs from our end. 

2. Compliance 

AI tools undergo the same scrutiny as any other software at TKC. This includes assessments to understand that they meet our requirements for encryptions, backups, compliance to GDPR and data location.   

We operate under an AI framework that ensures responsible use, and we’re audited annually against ISO27001:2002 and Cyber Essentials Plus. Learn more about our standards and accreditations here.

3. Quality & performance 

Our internal Quality Assurance team runs regular spot checks on AI outputs to minimise false positives or false negatives.  

While we don’t directly train the models ourselves, we build new ‘branches’ into our checking workflow so edge cases can be captured by bespoke AI checks, further reducing any inaccuracies. 

AI security solutions won’t replace you – but if you don’t adopt them, your competitors might

Adopting AI in physical security successfully shouldn’t mean reducing headcount – it should mean eliminating waste and boosting efficiency.  

For us, AI will never replace operators, officers, or control rooms, and that’s an ethos we advocate across the sector. The real risk lies in clinging to a “digital vs. human” mindset, which only fuels hesitation and stalls innovation. 

The reality is clear: in 3–5 years, AI-powered tools will be standard, not optional. Clients will expect real-time data, intelligent filtering, automated compliance, and smarter systems as the norm.  

Getting on board with bringing AI into your operations isn’t about being ‘trendy’ – it’s about defining the new baseline for great security. 

Taking your next step 

If you’re leading digital transformation, innovation or service delivery in physical security, ask yourself: is your operation ready for what’s next? Because the businesses that act now – building systems, asking the right questions, and embedding AI with purpose – will shape the future of security. 

For me, the path forward is clear: build the infrastructure, find the right use cases, and apply AI with care.