A case study for Accountable Autonomy at Airports

In recent years, an ever-growing number of AI-focused cybersecurity vulnerabilities have become commonplace in the aviation industry. In fact, according to a recent Axis Intelligence article, aviation cyberattacks surged 131% between 2022 and 2023, often the result of third-party vendors being regularly exploited.

Unlike traditional IT systems, AI models introduce novel vulnerabilities such as prompt injection, data poisoning, impersonation, and excessive agency that can compromise flight safety, operational integrity, and passenger security. These emerging risks demand an urgent reevaluation of cybersecurity frameworks to address AI-specific threat vectors before they can be exploited on a larger scale.

The Transportation Security Administration (TSA) has explicitly warned that AI systems such as large language models (LLMs), biometric scanners, and operational chatbots have become new attack surfaces that are not safeguarded by current aviation cybersecurity standards. Without integrated, unalterable verification and accountability mechanisms, these vulnerabilities undermine the trustworthiness of AI systems critical to airport operations and passenger safety.

For reference, according to a May 2025 article by generative AI security enterprise NeuralTrust, “A malicious actor could craft specific inputs to manipulate passenger-facing chatbots, performing unintended actions, such as retrieving unauthorized information, executing commands on backend systems it has access to, or generating inappropriate or harmful content attributed to the airline.”

Challenges of this nature often result in conflicting commands notifying airport personnel of operational inconsistencies and safety breaches, highlighting the critical gap in integrated, tamper-proof verification mechanisms that ensure trust and accountability across AI-driven systems.

The Multi-Stakeholder Complexity Problem

Airports represent some of the most complex technical ecosystems on Earth, supported by hundreds of systems such as data pipelines, communication protocols, and more, that are inherently managed by dozens of stakeholders.

These include airline operators and aircrafts, technology vendors, maintenance providers, regulators, security agencies, and others, all of which deploy proprietary AI and autonomous systems to ensure that airports remain safe and efficient at all times no matter the circumstances.

This complexity creates several risk types in airport environments, including:

● Opaque vendor systems: Airport operators struggle to carry out adversarial testing or validate log authenticity when vendors provide closed-source, proprietary solutions.

● Compromised audit trails: Event logs can be manipulated or concealed within protected vendor firewalls, eliminating reliable traceability after incidents.

● Mixed equipment operations: Heterogeneous system variability (consisting of AI models, autonomy levels, and communication protocols) can lead to compatibility issues, limited interoperability, and unexpected challenges for equipment operations.

● Human-AI friction: Ground personnel may maliciously prevent, modify, or override automated systems due to fears of job security or operational performance challenges, introducing safety risks that existing systems cannot verify or detect.

Some analysts warn that airport vendor risk is so severe that continuous maturity assessment and real-time monitoring have become baseline requirements. However, even with the implementation of such practices and the fact that traditional airport infrastructure has improved the operations and monitoring of airport applications and systems, it has not yet delivered data verifiability.

Exploring the Aviation AI Standards Landscape

Understanding the Importance of ISO/IEC 42001

As a whole, the AI standards landscape continues to make meaningful progress but is still susceptible to critical gaps related to AI integrity and cybersecurity.

Introduced in December 2023 as the first certified framework for responsible AI management, ISO/IEC 42001 is specifically designed to standardize processes for governance, data quality, and risk management for AI applicability across industries. Nonetheless, it is not able to verify that AI outputs themselves are authentic and tamper proof.

In February 2025, Changi Airport Group became the world’s first airport to achieve ISO/IEC 42001 certification, covering five AI-driven commercial and passenger applications including customer lifetime value forecasting, product recommendations, propensity prediction, generative AI search, and passenger transfer estimation.

In addition to Changi Airport Group’s recent certification, KPMG Australia and others have already obtained ISO 42001 certification, positioning the compliance framework as the standard for global AI regulation, specifically as the go-to standard supporting the EU AI Act.

When asked about the impact of ISO 42001 certification in the airline industry, the CEO of the British Standards Institution (BSI), Susan Taylor Martin, noted that, “AI is a transformational technology. For it to be a powerful force for good, trust is critical. The publication of the first international AI management system standard is an important step in empowering organizations to responsibly manage the technology, which in turn offers the opportunity to harness AI to accelerate progress towards a better future and a sustainable world.”

Explaining the Impact of the SAE G-34 & EUROCAE WG-114 Standards

Aviation-specific work by SAE International’s G-34 committee and EUROCAE’s WG-114 task force consisting of more than 600 industry-specific participants including Airbus, Boeing, Federal Aviation Administration (FAA), European Union Aviation Safety Agency (EASA), National Aeronautics and Space Administration (NASA), Thales, Honeywell, and several major Original Equipment Manufacturers (OEMs) will soon produce ED-324, the first standard upholding the development, certification, and approval of AI solutions in the aeronautics industry.

Specifically, the ED-324 standard will be focused on Airborne and Air Traffic Management (ATM) domains, covering machine learning (ML) data management, implementation and verification, system safety, and human systems integration, among others.

Existing security and cybersecurity-threat standards have made it challenging to adhere to compliance and regulatory requirements for numerous application-specific vulnerabilities related to safety. The newly developed ED-324 framework will help address the industry’s lackluster standards that have increasingly been susceptible to AI- and ML-specific cybersecurity threats like adversarial attacks and breaches and data and model manipulation.

Regulatory Momentum for AI Verification Frameworks

In recent months, regulatory momentum in AI aviation has continued to gain traction towards the development of more responsible AI systems in the aviation industry.

For example, the July 2024 FAA Roadmap for AI Safety Assurance report emphasizes risk-based safety assurance, incremental deployment, and continuous monitoring of AI learning systems. While the EU AI Act, which was initiated as law in August of last year, categorizes aviation AI as a high risk sector in dire need of stringent oversight, transparency, and compliance with ISO risk management standards.

In addition, many recent International Civil Aviation Organization (ICAO) workshops on AI in safety data analysis signal a worldwide movement among regulators toward unified frameworks for verification and accountability.

The Cost of Failure: Why Airports Can’t Afford Unverifiable AI

According to a June 2025 Markets and Markets report, the AI aviation market is projected to reach $4.86 billion by 2030 (20–23% CAGR), with autonomous systems increasingly becoming pivotal to streamlining operational safety, route optimization, and sustainability.

As outlined in a recent Air52 aviation article, autonomous ground handling has the potential to reduce ground damage costs by ~42% and typically improves turnaround efficiency by ~15%, but only if safety and accountability can be rigorously verified.

To ensure the safety of passengers and airport and flight staff, it is critically important that airports operate as safely as possible. Unfortunately, algorithmic failures can

sometimes result in aircraft collisions, passenger injuries, regulatory fines, while significantly harming organizational standing.

Fortunately, AI-driven predictive maintenance minimizes aircraft downtown and potential failures before they occur. That said, if AI models are tampered with and deliver incorrect results (such as false positives/negatives), the consequences can be disastrous.

According to GlobalData enterprise practice director Rena Bhattacharya, “No business wants to make headlines because they have used or misused customer or employee information in a way that causes harm. Reputational damage is not the only potential outcome and hefty fines and legal action could ensue.”

Clearly, airports need a way to ensure data is not only correct, but cryptographically verifiable and automatically executed at the blink of an eye across a plethora of systems.

Real-World Use Case: Winter Operations & Deicing

The winter months pose significant challenges related to airport safety and efficiency. In most regions in the Northern Hemisphere, snow, ice, and poor visibility create complicated multi-party coordination issues spanning deicing, aircraft guidance, ground safety, and more.

Thankfully, computer vision AI systems (such as those developed by JCAII) now deliver improved guidance for aircrafts when deicing, helping increase both safety and workflow productivity.

As a whole, these systems leverage AI-enabled real-time monitoring to track and oversee deicing operations, optimize ground traffic, and prevent safety incidences. Additionally, AI-focused systems can improve decision times for aircrafts by ~1 minute per flight, saving more than 140 hours of ground time per year (and $1.2M), for example, at Montreal-Trudeau airport alone, while reducing glycol consumption by 3–10%.

Airport authorities, airline deicing operations, third-party service providers, pilot groups, aircraft guidance systems, and AI vendors all contribute to establishing and meeting deicing Service Legal Agreements (SLAs). That said, there is often a lack of clear, integrated responsibility and enforceability across these entities, leading to gaps in accountability amongst stakeholders.

In addition, various interferences are commonplace in the AI aviation sector and can include:

● Alteration of logs via service providers that mask SLA breaches

● AI recommendations are sometimes unverifiably overridden by ground personnel

● Competing vendors may challenge accountability related delays or safety-specific incidents

● Regulatory enquiries, insurance and business audits often lack tamper-poor forensic record-keeping processes that support fair investigations

An April 2025 article by Synaptic Aviation highlights the effectiveness of AI systems in supporting harsh winter weather operations at airports as follows, “By integrating AI technologies into the operation of Passenger Loading Bridges at airports, operators can improve visibility and enhance overall safety and efficiency. AI systems can offer real-time analysis of environmental conditions, optimize operations for better performance in snowy weather, and ensure safer interactions between the PLB, aircraft, and ground personnel.”

Although digital transformation improves visibility in winter airport operations and aircraft deicing processes, a critical challenge remains: the absence of cryptographic data verifiability.

Without immutable verification integrated into AI airport systems, there is no way to ensure that logs and record data accurately reflect what truly occurred in a given circumstance. This gap enables potential tampering, unverifiable overrides of AI recommendations, disputes over stakeholder accountability, ultimately undermining the safety and efficiency gains AI systems aim to provide during complex winter operations.

The solution: verifiable autonomy via zero knowledge and Proof of Inference (PoI).

The Solution: Auditable Autonomy via Proof of Inference™

In the real-world, humans increasingly rely on cryptographic signatures and biometric systems like Face ID and Touch AI to safeguard accessibility to identity authentication systems and a myriad of online systems, meaning autonomous systems must leverage cryptographic verification to ensure operations are trustworthy.

Zero-knowledge proofs (ZKPs) enable the creation of cryptographic records that verify the correctness of AI outputs without revealing any underlying sensitive data or proprietary model details. Combined with Proof of Inference, ZKPs offer a robust, privacy-preserving mechanism to confirm decisions are authentic and untampered.

Inference Labs’ Proof of Inference™ framework ensures AI aviation system integrity by following a distinct step-by-step process. These include the following in order:

1. An autonomous AI system generates a prediction or initiates an operational action.
2. Zero-knowledge cryptography is applied to create a targeted verification of the AI’s output without exposing underlying private data, proprietary models, or sensitive vendor infrastructure.
3. For each AI decision, such as a aircraft guidance, ground equipment position, movement, hand off signal, or safety assessment, a cryptographic proof is produced. The proof serves as a non-reputable attestation of the system’s intentions and performance.
4. Vendors then publish these proofs to an open and neutral data store. In essence, proofs offer tamper-evident forensic accountability, disclosing no intellectual model property or sensitive data, while ensuring all actions are independently audible.

Building on the principles of Audible Autonomy and Model Informatics (the study of AI models), this framework is ushering in a new era of verification infrastructure for autonomous systems. In the bigger picture, these technologies ensure transparency, security, and trust in AI-driven operations, representing the fundamental verification infrastructure of the autonomous era.

Our solution delivers several key advantages for airport authorities and their stakeholders, encompassing:

● Coordinated multi-vendor oversight: Establishes transparent, sequential timelines based on proof of actions data to navigate complex stakeholder interactions.

● Regulatory alignment: Future proofed compliance with ISO 42001, EU AI Act, and FAA AI Safety Assurance requirements via fully auditable AI operational lifecycles.

● Forensic accountability: Facilitates dispute resolution, insurance claims, and safety investigations through immutable, verifiable audit trails.

● Confidentiality protection: Authenticates vendor AI activities without exposing proprietary algorithms or sensitive passenger information.

● Reliable cross-system operation: Ensures autonomous systems interoperate safely across heterogeneous standardizations and platforms.

Ultimately, by employing systems like those detailed above, verification becomes a computing primitive, enabling AI agents and robots to operate across networks, enterprises, and borders with verifiable certainty.

Market Opportunity & Industry Implications

Auditable Autonomy and airport AI is not only a critical safety and operational necessity, but a major market opportunity. Some prognostications hypothesize that autonomous transportation and logistics could add more than $400 billion to global GDP by 2035, while the civil aviation AI market is forecasted to grow from $4.9 billion in 2030 to $44.7 billion by 2035.

Regularity frameworks are rapidly evolving to support this transformation and ISO 42001 and related standardizations (ISO/IEC 238944 etc.) are becoming essential for AI deployment in regulated sectors, while mandates like the EU AI Act and FAA safety guidelines require explainability, auditability, and continuous monitoring of high-risk AI systems.

In addition, aeronautics standards such as SAE/EUROCAE’s ED-324 and others will set necessary certification baselines, meaning organizations without AI verification capabilities will risk non-compliance risk delays.

Early adopters (like Changi Airport) gain a critical competitive edge as industry leaders by demonstrating responsible AI governance, signaling trustworthiness to customers, regulators, and partners. As one expert puts it, “Companies deploying best practice guidelines might find themselves ahead of the competition. If people don’t trust AI, they won’t use it, no matter how advanced or beneficial it might be.”

In totality, Audible Autonomy via Proof of Inference, ZKPs, and related systems represent essential systemization frameworks for building trust, safety, and efficiency in modern aviation, turning compliance into a strategic advantage in a rapidly evolving industry. The future belongs to those who can prove their AI systems are not only innovative, but transparent, secure, and cryptographically verifiable.