Private security: how artificial intelligence is profoundly transforming the sector

Artificial intelligence is shifting the private security sector from a logic of constant surveillance to one of anticipation. Until recently, companies primarily protected shops, warehouses, or offices by deploying security guards, alarms, and cameras monitored from a control center. Now, software can simultaneously analyze hundreds of video feeds, detect unusual movement, intrusions, fires, abandoned objects, or gatherings, and then send an alert to an operator. France has approximately 300,000 people authorized to work in private security, according to the CNAPS (National Council for Private Security Activities), which gives an idea of ​​the scale of this ongoing transformation.

The most significant change, however, lies not in the cameras themselves, but in the integration of information. Images, alarms, access badges, thermal sensors, license plates, and incident logs can now all feed into a single platform. AI categorizes alerts according to their severity level and helps teams focus their attention on the most critical situations. In a warehouse, it can distinguish between an employee's normal presence and a nighttime intrusion. In a store, it can flag unusual behavior without waiting for an operator to notice it. The promise is considerable: faster intervention while reducing the time spent passively monitoring screens.

An economic revolution as much as a technological one

Large security companies are no longer simply selling guard hours, but contracts that combine agents, remote monitoring, software, maintenance, and threat intelligence. At Securitas, technology activities and security solutions saw real growth of 5% in 2025, as the group openly acknowledges its shift towards higher-margin digital services. In February 2026, the company also announced the acquisition of Liferaft, a platform that uses online information to detect threats targeting businesses. Private security is thus gradually expanding from building protection to the monitoring of digital and reputational risks.

The human agent doesn't disappear entirely, but their role changes. Systematic patrols can be partially replaced by smart cameras, drones, or sensors, while physical intervention remains the responsibility of professionals. This evolution enhances the value of operators capable of verifying an alert, understanding a computer system, and making a quick decision. It can also widen the gap between large corporations, able to invest in expensive platforms, and small businesses dependent on security contracts with low margins. AI thus risks further concentrating the market, while creating a massive need for training in a profession already facing workforce turnover.

The Olympic Games served as a laboratory

The Paris 2024 Games accelerated the acceptance of algorithmic video surveillance. The law had authorized, on an experimental basis, the automated analysis of images to detect certain predefined events during particularly high-profile events. The system was not intended to directly identify individuals, but rather to signal situations such as crowd movements, intrusions into restricted areas, or the presence of abandoned objects. Since March 2026, the experiment has been extended until December 31, 2027, according to the updated version of Article 10 of the law relating to the Olympic Games . This extension confirms that the Olympic exception has paved the way for a more permanent installation of these technologies.

The results, however, are far less spectacular than the marketing promises. The committee tasked with evaluating the Olympic trial drew up a mixed report, with performance varying depending on the location, camera angles, and the events being monitored. Dense crowds, poor lighting, or even a simple change in the environment can trigger unnecessary alerts. Each false positive requires an operator's attention and can divert attention from a real incident. While a claimed accuracy of 99% seems excellent, when applied to millions of crossings, even an error rate of 0,1% can produce a significant number of unjustified reports, as the European Commission points out.

This effectiveness requires safeguards.

The widespread adoption of these tools also raises a democratic question: to what extent can a private company observe, analyze, and categorize behavior in a public space? Images can reveal movements, habits, and relationships between individuals. Algorithms can also reproduce biases present in their training data or interpret perfectly ordinary behavior as suspicious. The European regulation on artificial intelligence specifically prohibits social scoring, certain forms of biometric categorization, and, except in strict exceptions, real-time remote biometric identification in public spaces. Furthermore, companies remain subject to the GDPR, the obligation to inform individuals, and the principle of proportionality.

AI, therefore, does not automatically make security more reliable. It makes it faster, more centralized, and potentially less expensive, but also more dependent on software whose operation can be difficult to control. The real game-changer lies in the sharing of responsibilities: when an algorithm misses an intrusion, triggers an unjustified intervention, or exposes personal data, the client, the security company, and the software provider can all pass the buck. The future of the sector will not be one of machines completely replacing agents, but rather one of hybrid surveillance in which AI detects, classifies, and recommends, while a trained professional must retain the final decision and be accountable for it.