This year, at the Munich Security Conference, AI is no longer discussed as a niche technical topic but as a core pillar of global stability, with dedicated sessions examining how autonomous systems, AI-enabled cyber operations, and intelligent agents compress decision-making time and blur traditional deterrence boundaries.
I had a keynote for Chief Security Officers (CSO) and focused on what AI can and can’t do to bridge between shiny powerpoints and implementation reality.
AI is turning chief security officers (CSOs) who protect people into intelligence leaders rather than just managers of guards, cameras, and badges. To keep people safe in a world where cyber and physical risks are tightly intertwined, we need to rethink security beyond firewalls and malware and redesign how we monitor spaces, move people, and respond to threats in the physical world.
How AI changes the day-to-day of people-focused CSOs
- From passive monitoring to real-time intelligence. AI can continuously analyse video, access control logs, and sensor data to detect anomalies such as tailgating, loitering, unauthorised perimeter activity, or weapons, and surface only relevant events to operators. This allows a smaller team in a GSOC to oversee many more sites while dramatically increasing situational awareness.
- Earlier detection and prevention of violence. Computer vision and behaviour analytics can spot early indicators of workplace violence, intruders, or unsafe behaviour and trigger interventions before an incident escalates, shifting security from response to prevention.
- Reallocation of human effort. AI can take over routine patrol-like monitoring, false alarm filtering, and basic report drafting, freeing human officers for high-touch tasks such as de-escalation, executive protection, crisis decision-making, and stakeholder communication. This changes hiring profiles and training priorities for future security teams.
Why we must rethink physical security, not just cyber
- Cyber–physical convergence. Many physical incidents now start digitally: a hacked access control system, spoofed credentials, or manipulated building automation can open doors to intruders or disrupt critical environments. CSOs need AI literacy to understand how AI-enabled cyber tools can directly translate into physical harm.
- Data-driven risk management for people. AI can correlate HR data, travel schedules, social media monitoring, incident logs, and environmental signals to highlight elevated risks to specific locations, events, or executives, enabling targeted protective measures instead of generic, expensive over-guarding.
- Ethics, bias, and trust in public spaces. Using AI for facial recognition, behaviour analysis, or workplace safety carries real risks of bias, over-surveillance, and erosion of trust among employees and communities. CSOs must lead conversations with legal, HR, and worker representatives to ensure safeguards, transparency, and proportionality are built into every AI deployment.
The new AI literacy for CSOs
For chief security officers, “knowing AI” now means understanding how computer vision, predictive analytics, and LLM-based assistants actually work, what their blind spots are, and how adversaries can game them. It also means being able to read AI-generated risk dashboards critically, ask the right questions of vendors, and design security concepts where AI extends human judgment instead of silently replacing it.
The most valuable time for me was during breaks, when several CSOs approached me to describe the unshiny reality of AI implementation in Fortune 50 companies. Mostly, these leaders were confronted with cost-saving targets, as now “AI was able to replace all these humans.” As in other business functions, no one was explaining where this AI had to come from, how to manage the transition, or what the standards are for identifying best vendors, managing technology lifecycle practices, or training employees. In one case, a CSO asked his CEO whether he preferred to be protected by software rather than his usual bodyguards. You can imagine what he answered.
I was very encouraged by the interest in looking into cybersecurity practices and protections. I described how cybercriminals used large language models to attack, and the importance of engaging white hackers and researchers to investigate scenarios of such attacks. Human errors and negligence play a significant role in exposing sensitive information, and we must collectively raise awareness of how shadow AI might undermine corporate defences. I remember those days of the “bring your own device” bonanza, when CIOs struggled to introduce control mechanisms to protect corporate networks.
Finally, cybercriminals use traditional Machine Learning to induce damage. They train models to probe and map weaknesses in camera networks and access control systems, finding blind spots in video coverage or patterns in guard routines that a human attacker might miss. Adversarial machine learning allows them to create specially crafted images, clothing patterns, or “patches” that cause person-detection algorithms to simply not see an intruder, effectively letting someone walk through a monitored area while appearing invisible to automated surveillance. At the same time, deepfake and voice-cloning techniques built on ML let criminals impersonate executives, security staff, or emergency responders in phone calls, intercom systems, and authentication workflows, tricking people into opening doors, disabling alarms, or moving high-value individuals to less secure locations. Together, these tools don’t just attack networks—they quietly erode the reliability of the very systems that physical security teams depend on, turning trust in cameras, sensors, and identity checks into a potential liability.
If you work in physical security and want a sparring partner to sanity-check which efficiencies really make sense when your team is given “cost savings through AI” targets, or you’d like to rethink the right blend of humans and technology to optimise your security operations, feel free to get in touch.