What Does Regulating AI Mean for Aviation Security
Since Artificial Intelligence has moved from experimentation into operational environments like airports and security authorities, it is important to consider not only what AI can do but how its use must be governed, monitored, and integrated into the existing aviation security frameworks.
AI is becoming part of aviation security with a prime example being the emergence of Automated Prohibited Items Detections (APIDS) algorithms and their use in European and U.S. airports. Both Amsterdam Airport Schiphol (AMS) and Frankfurt Airport (FRA) have deployed APIDS on checkpoint screening systems in Europe, and the United States Transportation Security Administration (TSA) is conducting operational testing and looking to pursue image on alarm only (IOAO) near-term (according to Administrator Cummins at a recent ACI PS&S Conference).
This system shows a shift in airport security: AI is not simply being tested as a future or “new” technology, but it is being put directly within the passenger and baggage screening processes. Therefore, the regulatory challenge is no longer a theory. When an AI system becomes part of the security chain, many questions on reliability, automated decision and human oversight, cybersecurity and system performance turn into operational concerns. A mistake, directly or indirectly due to the AI system’s assistance, may allow a prohibited item to pass through screening, or a false positive can lead to additional searches, delays or interventions. However, rigorous testing has been performed to ensure that these capabilities are equal to or greater than a human performing the same activity.
The EU AI Act provides a framework for addressing AI’s use in national security more broadly. Through these discussions, we have learned that not every AI system used at an airport should automatically be classified as “high risk”. To be more precise, classification depends on the intended purpose and the specific use of the system, which is an important distinction for aviation where AI may be used for everything from passenger flow management to biometric identification and security screening. Thus, an AI system for estimation of crowd levels at a gate raises different regulatory questions than an AI system for biometric identification.
Biometrics is likely to remain one of the most sensitive topics in aviation as the interaction between the AI Act and the EU’s General Data Protection Regulation (GDPR) continues to require further clarification. Airports and airline’s eagerness to use facial recognition and other biometric technologies to facilitate passenger processing, identity verification and access control is as it risk of being further curtailed by the lack of a dedicated risk-based approach to the use of biometrics in aviation. In addition, the AI Act is becoming connected to existing aviation security legislation. Regulation (EC) No 300/2008 has been amended to ensure that when technical specifications are made for the approval and use of aviation security equipment involving AI, the relevant AI Act requirements for high-risk systems are considered [3]. Therefore, the regulation of AI is becoming part of the existing aviation security framework instead of operating apart from it. EASA is working on the implementation of these requirements. NPA 2025-07 indeed proposes specifications and means of compliance for AI trustworthiness in aviation, including areas like AI assurance and human factors [4]. For operators, this means that having an AI system that performs well in a vendor demonstration is no longer enough. Procurement will need to require evidence, such as independent test results, documentation of how the model was trained and validated, and proof of compliance with AI Act requirements, that the system works safely and securely under real operating conditions. Deployment will also need ongoing checks, since an AI system’s performance can change over time with software updates, new threat types or shifts in passenger and baggage patterns.
Cybersecurity adds another layer of complexity. While aviation security focuses on preventing unauthorized access, prohibited items and acts of unlawful interference, AI presents us with a new way to attack, manipulating the system itself. An attacker could corrupt data used to train a detection model (data poisoning) so it learns to overlook threats. Researchers have already shown how X-ray screening algorithms can be fooled by deliberately designed objects [5]. Alternatively, attackers could tamper with the model or its software through a compromised vendor update, the same route used in the 2020 Solar Winds attack, where a malicious code was hidden in a routine software update installed by thousands of organizations [6].
Therefore, a security system that can be technically affected by an attacker creates different issues and needs appropriate testing and monitoring. Operators should therefore begin by understanding where AI is already being used across the security environment and assessing each application according to its purpose, regulatory classification, human-oversight requirements and security implications. This includes screening, biometric identification and verification, surveillance, access control, drone detection and security operations.
The immediate challenge is to ensure that AI does not become a separate technology project disconnected from existing security procedures. Its output will influence the security personnel’s attention, or how passengers move through checkpoints, and how potential threats are identified. So, technology needs to be assessed aspart of the wider security process.
AI is becoming part of the airport security infrastructure. Now, the question for European aviation is whether the regulatory, operational and human frameworks surrounding it can develop at the same pace.
LAM LHA is a capable partner in bridging technological constraints, regulatory frameworks, and operational realities, enabling the aforementioned stakeholders to move from reactive responses to coordinated, immediate action.
Get in touch.
