The Intelligent Airport
Full Text / Article Transcript
The Intelligent Airport: Artificial Intelligence as the Next Layer of Airport Safety Management
Albert N. Clark
Independent Author
Published: September 3, 2026
ASX Research Journal and Database
ISSN 3068-3351 (Online)
Place of Publication: Cadiz City, Philippines
Publisher: ASXResearch.org
Author Note
Albert N. Clark
Department of Aerospace Sciences, ASXResearch.org
ORCID iD: https://orcid.org/0009-0002-7348-4395
The author reports no conflicts of interest.
Correspondence concerning this article should be addressed to Albert N. Clark, Email: [email protected]
Abstract
Artificial intelligence is emerging as a potential new layer of airport Safety Management Systems by integrating operational, environmental, infrastructure, personnel, and surveillance data into a continuously updated assessment of airport risk. Rather than treating runway incursions, foreign object debris, wildlife hazards, pavement degradation, weather, staffing, fatigue, emergency response, construction, maintenance, and ground movement as isolated safety domains, the intelligent airport concept combines them to identify hazardous interactions that may remain invisible within traditional departmental structures. Current advances in machine learning, computer vision, surface surveillance, predictive maintenance, digital twins, and human-AI decision support demonstrate that many of the technological components already exist, while manufacturers, the Federal Aviation Administration, military organizations, and research institutions are progressively developing related capabilities. However, the transition from fragmented AI applications to an integrated airport-wide safety intelligence layer introduces substantial challenges involving explainability, certification, cybersecurity, privacy, employee monitoring, legal liability, public-sector governance, and human authority. This article argues that the most effective future model is not an airport controlled by artificial intelligence, but an airport whose Safety Management System is augmented by AI capable of continuously recognizing combinations of weak safety signals before they converge into an accident sequence. Such an architecture could transform airport safety from primarily reactive and periodically predictive risk management into persistent, system-wide anticipatory awareness.
Keywords: intelligent airports, artificial intelligence, Safety Management Systems
The Intelligent Airport: Artificial Intelligence as the Next Layer of Airport Safety Management
The intelligent airport is not really a story about replacing an airport manager, controller, firefighter, wildlife biologist, mechanic, security officer, or safety manager with an algorithm. It is the logical next stage in a transformation aviation began decades ago when safety moved from investigating accidents after the fact toward continuously identifying hazards before those hazards became accidents. Airports progressively accumulated digital surveillance, weather feeds, maintenance databases, airport operational databases, surface-movement systems, electronic NOTAM information, vehicle tracking, wildlife reports, inspection records, employee schedules, and formal Safety Management Systems, yet most of those systems were built as specialized islands. The revolutionary step is not simply applying artificial intelligence to each island; it is allowing an AI safety layer to examine the relationships among them. AlMarri, Bahroun, and Hassan (2026), reviewing 81 studies concerning AI-enabled airport risk management, reached a strikingly similar conclusion: airport AI research is already strong in individual functions such as FOD detection, pavement monitoring, runway prediction, weather forecasting, and surface-risk assessment, but remains fragmented and weakly integrated into Safety Management System governance. Yiu, Li, Ng, Chi, and Schiefele (2026) likewise found across 175 aviation-safety studies that AI has moved rapidly into accident analysis, human factors, operational safety, computer vision, and decision support while trustworthy deployment, explainability, certification, and human-AI teaming remain unresolved. In other words, most of the organs of the intelligent airport already exist; what aviation has not yet built at scale is the nervous system connecting them.
Figure 1
The Intelligent Airport: Artificial Intelligence as an Integrated Safety Management Layer Across Airport Operations and Risk Domains.
Note. Click image for full-size image or click here.
That distinction matters because an airport SMS is fundamentally a system for recognizing relationships. A cracked pavement panel, an understaffed maintenance shift, an approaching thunderstorm, an unusual wildlife concentration, a closed taxiway, a fatigued operations crew, an ARFF vehicle temporarily unavailable, and a construction NOTAM may each be acceptable when assessed independently; their simultaneous occurrence can create a completely different risk state. Traditional SMS depends heavily on people noticing those combinations, reporting them, bringing the right departments together, and conducting Safety Risk Management. Artificial intelligence offers the possibility of continuously performing that cross-domain correlation without abolishing human authority. Sharma, Wheeler, Chauhan, and Kelly (2026) demonstrated that supervised machine-learning techniques can extract predictive information from large aviation safety datasets and argued that such methods can strengthen the proactive and predictive foundation of SMS. That concept can be extended from accident records into airport operations: instead of merely asking whether an individual hazard exceeds a threshold, an intelligent airport could estimate whether multiple individually tolerable conditions are converging toward an intolerable system state. This is where AI becomes more than automation. The airport becomes a continuously observed socio-technical system whose safety condition can be reassessed every few seconds as weather, traffic, personnel, equipment, infrastructure, and operational restrictions change. Importantly, the FAA’s airport SMS framework already provides the governance structure into which this capability could fit; certain Part 139 airports are now required to develop and implement SMS, but the FAA explicitly does not mandate or approve any particular SMS software package.
The runway and taxiway system is probably the most mature laboratory for this idea because surface safety already generates precisely the kind of high-frequency, multi-variable data on which predictive AI thrives. Song, Cho, Tessitore, Gurcsik, and Ceylan (2018) demonstrated that runway-incursion risk can be approached statistically by combining diverse operational variables and explicitly quantifying uncertainty rather than waiting for an incursion to occur. More recent work has become markedly more sophisticated. Yuan, Fang, Chen, and Liu (2025) combined graph neural networks with long short-term memory models to represent airport taxiway topology and temporal traffic behavior, allowing future surface-conflict conditions to be predicted rather than merely detected. Tian, Li, Zhou, Sun, and Shi (2026) subsequently developed a self-attention gated recurrent unit framework capable of forecasting changing airport conflict hotspots, reporting materially lower prediction error than benchmark models and showing that the location and intensity of risk vary over time. This is a profound shift in runway safety philosophy. ASDE-X, ADS-B, multilateration, vehicle transponders, surface radar, and runway-status systems traditionally tell operators what is happening now; AI can begin estimating what is becoming dangerous next. The FAA is already strengthening the sensing foundation by expanding airport-vehicle transponder use, while its Runway Status Lights automatically process surface-surveillance data to warn pilots and vehicle operators of conflicts. An intelligent-airport layer could sit above these technologies and ask the larger question: given the next ten minutes of aircraft movement, vehicle movement, visibility, runway configuration, construction restrictions, controller workload, and historical conflict patterns, where is the next unacceptable surface risk likely to emerge?
The same architecture becomes enormously powerful when it looks away from aircraft trajectories and begins examining the physical airport. Computer vision can continuously inspect runways, taxiways, aprons, fences, lighting, markings, aircraft stands, and service roads. Mo, Wang, Hong, Chu, Li, and Xia (2024) demonstrated real-time deep-learning detection of small-scale foreign object debris using visible and infrared imagery, illustrating how a formerly labor-intensive inspection problem can become persistent machine observation. The FAA itself has tested small-unmanned-aircraft and AI/ML-based FOD detection at Cape May County Airport and Atlantic City International Airport, achieving a reported 96% detection rate for specified FOD items in its proof-of-concept work while also identifying false-positive, processing-time, weather, and low-light limitations that still require development. Pavement offers another avenue: Clemmensen and Wang (2024) showed that machine-learning models can predict airfield pavement condition and support life-cycle maintenance decisions, while Qi, Xie, and Shi (2026) integrated a neural network with fuzzy decision analysis to prioritize concrete-airport pavement maintenance while preserving interpretability. The same sensor fusion can be extended to lighting failures, drainage problems, fence incursions, standing water, rubber contamination, vegetation, wildlife movement, snow or debris, and deteriorating markings. Wildlife management is especially attractive because the relevant variables are not merely visual; species, season, rainfall, temperature, time of day, mowing, nearby land use, migration, waste handling, standing water, and recent strike history can be combined into a dynamic probability map. Instead of sending inspectors everywhere with equal urgency, the airport could continuously ask where physical degradation or environmental conditions are most likely to become an operational hazard next.
Weather and capacity management reveal why the proposed system must reason across departmental boundaries rather than optimize isolated functions. Wang and Zhang (2021) demonstrated a deep-learning method that used gridded weather forecasts to predict runway configurations and airport acceptance rates across a multi-airport system, capturing dependencies that point observations alone could miss. Andy, Alam, Lilith, and Piplani (2024) went further by applying deep reinforcement learning to runway-configuration management at Philadelphia International Airport, using real operational, meteorological, and traffic data and reporting substantial reductions in modeled delayed flights. The crucial safety implication is not that an AI should arbitrarily select runways; it is that the system can calculate consequences across a much larger state space than a human team can mentally hold at once. A runway configuration change affects taxi patterns, crossings, runway occupancy, workload, gate conflicts, deicing flow, ARFF positioning, vehicle routes, wake separation, and exposure to weather. The intelligent airport would therefore treat weather not as a separate meteorological feed but as a variable propagating through the entire operational model. A thunderstorm would change more than the arrival rate: it could alter ramp exposure, personnel availability, lightning restrictions, surface friction, drainage, gate occupancy, fuel servicing, emergency access, and maintenance priorities simultaneously. The AI layer becomes valuable precisely because aviation accidents rarely respect organizational charts. It can search for interactions between systems whose human owners may work in different rooms, report to different departments, use different software, and never realize that their individual problems have just combined into one dangerous airport-level condition.
The human side may ultimately be more consequential than the machinery. Airport safety depends upon controllers, operations specialists, mechanics, marshallers, tug drivers, baggage personnel, fuelers, security officers, contractors, ARFF crews, dispatchers, snow teams, and managers whose performance changes with workload, fatigue, training, distraction, heat, overtime, staffing shortages, and task saturation. Xin, Lim, Hsieh, Chen, and Dong (2025) reviewed 105 peer-reviewed studies concerning AI and fatigue in air traffic control and concluded that AI can combine behavioral, physiological, and other indicators to strengthen fatigue-risk management, while simultaneously emphasizing the need for a human-centered approach. An intelligent airport could therefore incorporate staffing and workload without becoming a dystopian employee-surveillance machine. For example, it might recognize that an understaffed ramp, reduced ARFF availability, a heavy arrival bank, high heat index, a disabled taxiway, and deteriorating visibility collectively warrant an elevated operational risk state. Emergency response could similarly become predictive: AI could continuously recompute ARFF access routes based on closed pavement, aircraft positions, construction, traffic, wind, fuel hazards, and vehicle availability, giving commanders decision support before an emergency occurs. But personnel prediction is exactly where technical capability collides with ethics. Fatigue estimation, facial analysis, biometrics, productivity data, medical information, and worker monitoring create privacy, employment, labor, discrimination, and due-process questions. The intelligent airport therefore needs an architectural principle as important as predictive accuracy: the machine should assess operational risk without casually converting every employee into a permanently scored biological sensor.
Industry is already moving toward pieces of this architecture, and one manufacturer now comes remarkably close to the concept described here. Among the publicly documented systems I found, ADB SAFEGATE’s CORTEX platform is probably the clearest commercial precursor to an airport-wide AI safety layer: the company describes one software family spanning airfield, apron, tower, weather, and service functions with a common data model, security framework, and AI foundation, while its “Airside Intelligence” concept is explicitly intended to turn airside data into foresight. Its apron products already include AI-assisted FOD detection, stand and gate awareness, predictive operational functions, and surface safety nets. Saab’s A-SMGCS provides another major piece of the puzzle through integrated surveillance, vehicle tracking, routing, guidance, and airport safety-support algorithms, and the Philadelphia reinforcement-learning research cited above was conducted under the Saab–Nanyang Technological University Joint Laboratory. Honeywell markets a unified airside-operations platform combining visual guidance, lighting control, traffic management, and situational awareness, while SITA provides shared real-time airport operational data and collaborative decision tools. SITA’s 2025 industry survey reported that 73% of airports were investing in AI and that nearly half planned greater cross-silo data streaming or synchronization by 2027. None of these public offerings yet appears to implement the complete airport-wide SMS intelligence envisioned here—simultaneously reasoning over runway incursions, FOD, wildlife, infrastructure, staffing, fatigue, ARFF, security, weather, NOTAMs, construction, and formal SMS risk controls—but the commercial trajectory is unmistakable. The pieces are ceasing to be isolated products and are beginning to look like a platform.
The FAA is not standing outside this transition. Its 2024 Roadmap for Artificial Intelligence Safety Assurance explicitly distinguishes between assuring the safety of AI and using AI to improve safety, and in 2026 the agency established AI/machine learning as a defined technical discipline supporting policy, guidance, training, validation, and international coordination. At the airport level, the 2023 Part 139 SMS rule creates an especially important institutional bridge because qualifying certificated airports must move toward systematic hazard identification, safety-risk management, assurance, and promotion rather than treating AI as an IT experiment. The FAA’s research branch has already investigated AI-assisted FOD detection, and the agency continues investing in surface surveillance, vehicle visibility, and automated runway protections. The NTSB occupies a fundamentally different position: it does not certify airport AI or operate airport safety systems; it investigates accidents and issues recommendations. Its runway-safety recommendations nevertheless create strong demand signals for better surface detection, alerts, procedures, and SMS-based analysis, including a recommendation that certain Part 139 airports without appropriate surface-detection systems receive equipment capable of tracking aircraft and providing controllers visual and aural cues. Thus, the FAA is likely to become the principal federal gatekeeper for operational acceptance and safety assurance, while the NTSB will become one of the most important sources of failure evidence against which intelligent-airport systems are judged. If an AI claims it can identify dangerous combinations, investigators will eventually ask a brutal but necessary question after an accident: did the system possess the relevant information, and if so, why did nobody act on it?
The military is pursuing many of the underlying technologies, although I found no public evidence of a Department of Defense or DARPA program attempting to create this exact airport-wide civil-style SMS intelligence layer. U.S. military airfields have obvious reasons to automate damage assessment, runway recovery, weather prediction, logistics, maintenance, autonomous surveillance, and operational decision support; Air Force organizations have experimented with rapid airfield-damage sensing and continue exercising the integration of damage assessment, debris clearance, engineering, and rapid runway restoration. DARPA, meanwhile, is deeply involved in aviation AI through programs such as Air Combat Evolution and Artificial Intelligence Reinforcements, but those efforts are aimed primarily at autonomous tactical aircraft, human-machine teaming, and distributed combat behavior rather than airport SMS. That distinction should not be blurred simply because all of them contain AI. DARPA’s work could eventually contribute assurance methods, uncertainty management, human-machine interfaces, or autonomous-agent architectures to airport applications, but there is no basis at present for calling DARPA the developer of an “intelligent airport” program. The same caution applies to Elon Musk. Musk is highly active in artificial intelligence, autonomous systems, spaceflight, and current AI-governance debates, but I found no credible public program from Musk, xAI, Tesla, or SpaceX developing an integrated airport safety-management intelligence system of the kind examined here. His companies may possess technologies relevant to the problem; that is not the same thing as actually tackling the problem. Surgical accuracy requires leaving that answer at “not presently demonstrated,” rather than inventing a connection because the names AI, autonomy, aviation, and Musk happen to occupy the same technological universe.
The legal ramifications become fascinating the moment AI crosses the line between observing a hazard and recommending an operational action. An airport operator cannot plausibly delegate its regulatory responsibility to a neural network and then blame the algorithm when something goes wrong. The FAA’s own Part 139 guidance makes clear that SMS responsibility remains with the airport certificate holder, and the agency specifically states that it does not approve airport SMS software. Consequently, an intelligent-airport system would need extensive records showing what data it received, what model version was operating, what risk it calculated, what uncertainty accompanied the result, what recommendation it produced, who received it, whether a human accepted or rejected it, and what subsequently occurred. Those records could become central evidence in accident investigations, negligence litigation, product-liability disputes, insurance claims, employee actions, regulatory enforcement, and discovery. Explainability therefore stops being an academic luxury and becomes a legal-defense mechanism. A black-box model announcing “close Runway 18” is far harder to defend than a system showing that braking reports deteriorated, rainfall intensity increased, two prior hydroplaning precursors emerged, maintenance data indicated marginal grooving, and the modeled excursion probability crossed a documented SMS threshold. NIST’s AI Risk Management Framework is voluntary rather than aviation law, but its emphasis on governance, validity, reliability, safety, resilience, accountability, transparency, explainability, privacy, and continuous lifecycle risk management maps extraordinarily well onto what airport AI will require. The likely regulatory endpoint is not certification of “intelligence” in the abstract; it is rigorous assurance of defined functions, bounded authority, training data, failure modes, cybersecurity, human oversight, and operational consequences.
Civil ramifications extend well beyond the FAA because many American airports are owned or operated by cities, counties, states, port authorities, or other public entities. That makes the intelligent airport simultaneously an aviation system and a piece of public infrastructure. Local governing boards may control procurement; state law may govern contracting, records retention, biometric information, employee monitoring, privacy, cybersecurity, tort exposure, and public access to government records; municipal emergency services may participate in airport response; county planners may control land uses affecting wildlife; state transportation agencies may fund pavement or infrastructure; and local police may contribute security information. An AI layer that consumes all of those streams can therefore create questions that an ordinary airport database never raised. Should a safety model have access to security-camera facial recognition? Can employee fatigue information be used for discipline? Are proprietary model outputs public records when produced for a municipal airport? Who owns data contributed by an airline, tenant, contractor, county fire department, and FAA system? What happens when an algorithm disproportionately flags one employee group, neighborhood, contractor, or category of traveler? What cybersecurity standard applies when compromising a single platform could corrupt wildlife warnings, maintenance priorities, surface-risk predictions, and emergency routing simultaneously? The answer cannot be that “the AI decided.” Local and state governments will need explicit data-governance agreements, procurement standards, audit rights, retention policies, human-review requirements, cybersecurity controls, vendor-liability provisions, and clear boundaries around surveillance. This is also why the intelligent airport should initially behave as an SMS decision-support layer rather than a sovereign decision maker: combining data across organizational boundaries is one of its greatest safety advantages and simultaneously one of its greatest civil-governance risks.
Where this ultimately goes is considerably more interesting than another dashboard filled with colored risk boxes. The mature intelligent airport could become a continuously updated digital safety model of itself: every aircraft, vehicle, runway, taxiway, gate, weather cell, construction project, NOTAM, lighting circuit, pavement segment, wildlife report, staffing state, emergency asset, and unresolved SMS hazard represented within a common temporal model. Machine learning would identify patterns; computer vision would observe the physical environment; graph networks would model relationships; language models could interrogate reports, NOTAMs, maintenance narratives, and procedures; digital twins could test interventions before they are implemented; and probabilistic models would express uncertainty rather than pretending to possess certainty. The decisive advance would be causal and combinational reasoning—recognizing that nothing individually alarming is occurring while simultaneously recognizing that six modest degradations have just aligned into tomorrow’s accident. AlMarri, Bahroun, and Hassan (2026) argue that the field must now move from fragmented AI experiments toward interoperable, explainable, governance-ready systems embedded in SMS, while Yiu, Li, Ng, Chi, and Schiefele (2026) similarly identify explainability, human-AI teaming, diversified data, and assurance frameworks as critical to aviation AI’s next stage. That is exactly the threshold on which the intelligent airport now stands. The future airport will not be “run by AI.” The safer and far more powerful possibility is an airport that can finally perceive itself as one system—and can warn its humans when separate pieces of that system are quietly beginning to assemble the conditions for an accident.
References
AlMarri, M., Bahroun, Z., & Hassan, N. M. (2026). Artificial intelligence for safety and resilience in airport transportation systems: A systematic review of operational, security, and environmental risks. Transportation Research Interdisciplinary Perspectives, 37, 101961. https://doi.org/10.1016/j.trip.2026.101961
Andy, L. J. G., Alam, S., Lilith, N., & Piplani, R. (2024). A deep reinforcement learning approach for runway configuration management: A case study for Philadelphia International Airport. Journal of Air Transport Management, 120, 102672. https://doi.org/10.1016/j.jairtraman.2024.102672
Clemmensen, A., & Wang, H. (2024). Airfield pavement condition prediction with machine learning models for life-cycle cost analysis. International Journal of Pavement Engineering, 25(1), 2322529. https://doi.org/10.1080/10298436.2024.2322529
Mo, Y., Wang, L., Hong, W., Chu, C., Li, P., & Xia, H. (2024). Small-scale foreign object debris detection using deep learning and dual light modes. Applied Sciences, 14(5), 2162. https://doi.org/10.3390/app14052162
Qi, L., Xie, Y., & Shi, X. (2026). Maintenance schedule optimization method of cement concrete airport pavement based on BP neural network and FAHP. Journal of Transportation Engineering, Part B: Pavements, 152(3), 04026037. https://doi.org/10.1061/JPEODX.PVENG-1883
Sharma, V., Wheeler, B., Chauhan, B., & Kelly, S. (2026). Data-driven decision-making in aviation Safety Management Systems: A supervised machine learning approach. Collegiate Aviation Review International, 44(1), 178–199.
Song, I., Cho, I., Tessitore, T., Gurcsik, T., & Ceylan, H. (2018). Data-driven prediction of runway incursions with uncertainty quantification. Journal of Computing in Civil Engineering, 32(2). https://doi.org/10.1061/(ASCE)CP.1943-5487.0000733
Tian, W., Li, Y., Zhou, X., Sun, J., & Shi, X. (2026). Time-series risk forecasting of airport conflict hotspot with SA-GRU. Journal of Air Transport Management, 133, 102977. https://doi.org/10.1016/j.jairtraman.2026.102977
Wang, Y., & Zhang, Y. (2021). Prediction of runway configurations and airport acceptance rates for multi-airport system using gridded weather forecast. Transportation Research Part C: Emerging Technologies, 125, 103049. https://doi.org/10.1016/j.trc.2021.103049
Xin, S., Lim, K. Y. H., Hsieh, M.-H., Chen, C.-H., & Dong, L. (2025). Managing the fatigue frontier: AI applications in air traffic control operations. Advanced Engineering Informatics, 68, 103660. https://doi.org/10.1016/j.aei.2025.103660
Yiu, C. Y., Li, W.-C., Ng, K. K. H., Chi, C.-F., & Schiefele, J. (2026). Enhancing aviation safety with artificial intelligence: A systematic literature review on recent advances, challenges and future perspectives. Advanced Engineering Informatics, 71, 104378. https://doi.org/10.1016/j.aei.2026.104378
Yuan, L., Fang, D., Chen, H., & Liu, J. (2025). Prediction of airport surface potential conflict based on GNN-LSTM. IET Intelligent Transport Systems, 19(1), e12611. https://doi.org/10.1049/itr2.12611