By Professor Rocky Scopelliti
Futurologist and Author of The Conscious Code: From Artificial Intelligence to Artificial Consciousness
Artificial intelligence is changing a fundamental relationship in security.
For decades, security technology largely helped humans see.
CCTV showed an operator what was happening. Alarms notified monitoring centres. Access-control systems recorded who entered. Analytics helped investigators understand what had occurred.
AI increasingly does something different.
It can detect, interpret, prioritise and recommend a response—and increasingly autonomous systems could begin initiating parts of that response themselves.
That means the next challenge for security professionals is not simply using AI to watch people, places and assets.
It is maintaining meaningful human oversight of the machines doing the watching.
From detection to security authority
Consider an AI-enabled video surveillance system.
At its simplest, it detects movement.
More sophisticated systems may classify behaviour, identify anomalies, assign risk or prioritise an event for human attention.
Connect that intelligence to access control, alarm monitoring, communications or other systems and its recommendations can begin influencing what happens in the physical world.
The technology has crossed an important boundary.
In my research for The Conscious Code: From Artificial Intelligence to Artificial Consciousness, I distinguish between machine capability and the authority humans delegate to machines.
AI does not need consciousness to exercise consequential operational agency.
For security professionals, the practical question is therefore not simply:
What can the AI detect?
It is:
What are we allowing the AI to do because of what it believes it detected?
The Security AI Authority Threshold
I propose thinking about security AI across five stages:
DETECT → INTERPRET → PRIORITISE → DIRECT → ACT
At Detect, AI identifies an event or signal.
At Interpret, it decides what that signal may mean.
At Prioritise, it determines which event deserves attention first.
At Direct, it recommends or issues an operational response.
At Act, it initiates something capable of creating a physical or consequential outcome.
The further AI moves along that chain, the stronger the requirements should become for human oversight, traceability and intervention.
That principle applies differently across NZSA’s membership.
Protective security: when the machine and officer disagree
Imagine AI dynamically allocating patrols according to predicted incidents.
The system directs a security officer towards one location.
But the officer observes something locally that the AI cannot see and believes another response is necessary.
Who has practical authority?
If ignoring the machine requires justification, affects performance metrics or is treated as non-compliance, AI advice can quietly become AI direction.
Frontline personnel should retain meaningful authority to challenge machine-generated instructions when their situational awareness tells them something the system does not know.
Their disagreement should also become intelligence.
Why did the human see something the machine missed?
Electronic security: detection is becoming interpretation
Electronic security presents another boundary.
AI-enabled CCTV and video analytics can increasingly distinguish between ordinary and anomalous behaviour.
That can dramatically improve security operations by helping humans manage enormous volumes of information.
But an anomaly is not necessarily a threat.
If machine interpretation automatically influences access, escalation or another physical response, the consequences of a false inference increase.
Security professionals therefore need to understand not merely system accuracy, but the authority attached to the inference.
Alarm monitoring: the event the machine did not escalate
AI could be enormously valuable in alarm monitoring.
It can potentially reduce false alarms, correlate multiple signals and help operators concentrate on events most likely to require intervention.
But prioritisation inevitably means some signals receive less attention.
Imagine an AI correctly filters thousands of routine events—and incorrectly deprioritises the one event that matters.
That creates what I call the Detection Gap: the interval between an intelligent system moving outside its intended operating boundary and a human recognising that it has done so.
The principle is simple:
You can’t stop what you can’t see.
Monitoring therefore needs to include the AI itself.
How quickly would operators recognise that its classifications or priorities had become abnormal?
Security consultants will increasingly govern machine authority
This may create an important new role for security consultants and integrators.
Clients will need advice not merely about which AI-enabled system performs best.
They will increasingly need answers to questions such as:
What can this system autonomously do?
Which other systems can it access?
What happens when its confidence falls?
When must a human approve an action?
How quickly can unexpected behaviour be detected?
Can autonomous authority be withdrawn?
And can we reconstruct what happened afterwards?
In The Conscious Code, I describe that final requirement as the Trace Test.
After a consequential machine action, can we determine what the system perceived, what it inferred, which authority it exercised, what action followed and where humans intervened?
Security’s next object of protection
AI could make security substantially more effective.
It can help identify threats earlier, reduce false alarms, improve resource allocation and allow security professionals to focus attention where it matters most.
The objective should not be less AI.
It should be deliberate machine authority.
Security professionals have spent generations designing systems to detect when people cross boundaries they should not cross.
Agentic AI introduces an intriguing reversal.
Increasingly, we may also need security systems capable of detecting when the machine itself crosses the boundary we gave it.
Because the defining question for the next generation of security may no longer simply be:
“What is the AI watching?”
It may be:
“Who is watching the AI?”
About Professor Rocky Scopelliti
Professor Rocky Scopelliti is an internationally recognised futurologist, best-selling author, international keynote speaker and Adjunct Professor whose work explores The Evolution of Intelligence—how artificial intelligence is transforming business, society and what it means to be human.
He is the General Editor of The Evolution of Intelligence Series, published by CRC Press/Taylor & Francis and author of Synthetic Souls, Perceptive Machines, and The Conscious Code, the first three books in the series.
Rocky advises governments, boards and global organisations on the implications of emerging technologies and delivers keynote presentations worldwide on the future of intelligence, leadership and humanity.
Learn more: RockyScopelliti.com

