Automated Systems, Certification Failure and the Liability Gap Above 30,000 Feet
From the 737 MAX to the eVTOL Certification Race: How Aviation Law Is Absorbing Algorithmic Flight
Introduction
In May 2026, LOT Polish Airlines became the first airline to take Boeing to a jury trial over the 737 MAX grounding.1 The airline sought approximately USD 153 million in damages arising from the worldwide grounding of the MAX fleet following the Lion Air and Ethiopian Airlines crashes that killed 346 people. The jury found that LOT had failed to substantiate its claims of purposeful misrepresentation regarding Boeing’s disclosure of the Maneuvering Characteristics Augmentation System.1 The verdict was a defence win for Boeing on the airline’s commercial claims, but it resolved nothing about the underlying automated-systems liability that produced the two crashes.
Publicly reported MAX-related DOJ resolutions alone exceed USD 3.6 billion, before taking account of confidential civil settlements. The federal criminal case was dismissed in November 2025 after the Department of Justice requested dismissal. As part of the resolution, Boeing agreed to pay or invest an additional USD 1.1 billion in fines, victim family compensation and internal safety measures.1 Separately, Boeing has resolved more than 90% of the individual civil complaints arising from the two crashes. The wrongful death trial of Michael Ryan is scheduled to commence on 3 August 2026.1 A jury awarded USD 49.5 million to the family of Samya Stumo, killed in the Ethiopian Airlines crash, in May 2026.1
The 737 MAX is not an AI case. MCAS was a deterministic automated system that responded to angle-of-attack sensor data according to pre-programmed logic. It did not learn or adapt. It executed its programming and its programming was fatally flawed. But the liability framework that the MAX litigation has constructed applies directly to the next generation of aviation AI: adaptive flight-management systems, AI-enabled predictive maintenance, machine-learning-based air traffic management and the autonomous flight capabilities being developed for the eVTOL air taxi sector. The allocation question is the same in each context: when an automated system fails and people die, how does the law distribute responsibility between the manufacturer, the certifier, the operator, the software supplier and the system integrator?
The Certification Gap: FAA and EASA
Aviation is the most heavily regulated transport sector. Aircraft certification requires demonstration of compliance with airworthiness standards through a process that can take years and cost hundreds of millions of dollars. The certification framework was designed for deterministic systems: mechanical and electronic components that behave predictably and can be tested exhaustively against failure-mode analysis.
AI-enabled systems strain this model. A machine-learning system whose behaviour depends on training data, operational context and later updates cannot be characterised in the same way as a conventional deterministic component. The European Union Aviation Safety Agency recognised this in its AI Roadmap and in November 2025 published Notice of Proposed Amendment 2025-07, the first regulatory proposal for AI trustworthiness in aviation.2 The NPA distinguishes between Level 1 AI (AI-based assistance, where the human retains full authority) and Level 2 AI (human-AI teaming, where the AI shares decision-making with the pilot).2 The consultation closed in March 2026, after an extension from the original February deadline, and a second NPA addressing domain-specific regulations is expected later in 2026.2
The FAA has taken a slower path. Its Roadmap for Artificial Intelligence Safety Assurance sets out principles for the safety assurance of AI in aircraft and aircraft operations, but has not produced binding certification standards for AI in flight-critical systems.2 SAE G-34/EUROCAE WG-114 is developing ARP6983/ED-324 as a recommended practice for AI/ML in aeronautical systems, but the first version is limited and does not solve certification of the highest-criticality adaptive systems.2 The current scope extends to Design Assurance Level C, corresponding to “major” failure conditions. It does not cover DAL-A (catastrophic failure) or DAL-B (hazardous failure), which govern the systems most likely to kill people.
The gap between what AI can do in aviation and what regulators can certify is widening. The technology is advancing more quickly than the regulatory framework that must validate it for flight.
The eVTOL Certification Race
Electric vertical takeoff and landing aircraft are the first new commercial aviation category in which highly automated flight-control software, autonomy roadmaps and certification constraints converge from the outset. The eVTOL sector has attracted approximately USD 13 billion in investment since 2019.3 Four concurrent certification applications are being processed by the FAA: Joby Aviation, Archer Aviation, Beta Technologies and Wisk Aero.3
In November 2025, Joby began power-on testing of the first FAA-conforming aircraft being built for Type Inspection Authorization testing. In March 2026, it began flight testing the first aircraft intended for FAA certification testing, with FAA pilots expected to conduct “for credit” TIA testing later in 2026.3 As of the first quarter of 2026, no US eVTOL manufacturer has received a full type certificate for commercial passenger operations.3
Lilium’s principal German operating subsidiaries filed for insolvency in late October 2024, and Lilium N.V. later authorised insolvency proceedings after failed fundraising efforts.3 The company had received EASA Design Organisation Approval in November 2023 but pursued a novel jet-propulsion architecture that fell outside any existing certification framework, resulting in extended timelines and unsustainable cash burn. Lilium’s failure demonstrates that the certification pathway is itself a commercial risk of the first order: a technically viable aircraft that runs out of capital before it completes certification cannot generate revenue.
The product liability implications for eVTOL are significant. These are new-category aircraft operating in urban environments over populated areas. The first passenger fatality will produce litigation in which the manufacturer, the battery supplier, the flight-control software provider and the operator will all be in play, while the FAA’s certification decision may form part of the factual and regulatory background. The automated flight-control systems that manage hover-to-cruise transition, motor failure redistribution and autonomous emergency landing are the systems most likely to be at the centre of that litigation.
AI in Maintenance and Predictive Failure
AI-enabled predictive maintenance is one of the earliest and most commercially advanced applications of machine learning in aviation. Airlines and MRO (maintenance, repair and overhaul) providers use AI systems to analyse engine sensor data, flight-data recorder outputs, component wear patterns and environmental exposure data to predict when parts will fail and schedule maintenance before failure occurs.
The hard issue for courts will arise when a predictive system fails to predict. If an AI maintenance system assesses a turbine blade as having adequate remaining life and the blade subsequently fails in flight, the causation analysis involves the manufacturer of the AI system, the airline that relied on its output, the MRO provider that implemented the AI-recommended maintenance schedule and the regulatory framework that permitted or required AI-driven maintenance planning.
The standard of care for maintenance decisions is well established in aviation law. Maintenance must comply with the manufacturer’s maintenance programme and the applicable airworthiness directives. AI-enabled predictive maintenance introduces a tension between approved maintenance intervals and tool-generated assessments of component condition. If an operator seeks to extend service life or alter inspection timing on the basis of AI analysis and the component subsequently fails, the critical question will be whether that decision sat within an approved maintenance or reliability programme. The operator’s deviation from published maintenance intervals will be a central issue in the litigation.
Drone Delivery and Airspace Liability
Drone delivery in the United States currently operates through a combination of Part 135 air-carrier certification, Part 107 small-UAS rules, exemptions and individual airspace authorisations for each delivery zone. Operators including Wing, UPS Flight Forward, Amazon Prime Air and Zipline have used the Part 135 pathway.4 FAA environmental and operational review materials for Zipline projects contemplate operations of up to 400 delivery flights per operating day from each site and, in some cases, 24-hour operations.4 The proposed Part 108 rule for routine beyond-visual-line-of-sight operations is intended to simplify the fragmented structure, but until it is finalised the regulatory path remains piecemeal.4
Liability for drone delivery incidents operates under conventional aviation and product liability principles. The drone operator will face the primary operational and regulatory exposure under its Part 135 authority. The manufacturer is liable for design and manufacturing defects. The question that has not been tested is the liability of the AI navigation system that determines the drone’s flight path, altitude, obstacle avoidance and delivery approach. If an autonomous delivery drone strikes a person or property, the liability allocation between the operator, the manufacturer, the navigation software provider and the airspace authority that approved the delivery zone will depend on whether the cause was a software defect, a sensor failure, an environmental condition outside the system’s design parameters or a regulatory failure in approving the operating zone. Any claim against the FAA or another airspace authority would face public-law and sovereign-immunity barriers, so the practical liability focus is likely to remain on the operator, manufacturer and software suppliers.
Strategic Outlook
Aviation’s regulatory culture of exhaustive certification and conservative adoption of new technology has slowed the integration of AI into flight-critical systems. This conservatism has so far helped avoid AI-related aviation disasters in flight-critical civil aviation systems. It has also created a growing disparity between what AI can do in aviation and what the certification framework permits.
The 737 MAX litigation established that automated-system design defects, combined with certification failures, can produce multi-billion-dollar exposure across regulatory enforcement and civil proceedings. The eVTOL certification race may produce the first highly automated new-category aircraft type certificates within the next two years, although the extent to which those certificates involve AI rather than conventional flight-control automation will vary by platform. The expansion of drone delivery to 24-hour, high-volume operations over populated areas is likely to produce significant drone-injury litigation.
Aviation law is not creating a new liability framework for AI. It is applying existing certification, product liability and carrier liability principles to increasingly autonomous systems. The question is whether those principles, designed for deterministic systems that behave predictably, can accommodate AI-enabled systems whose behaviour may be difficult to characterise fully in advance and which may fail in ways their designers did not anticipate.
Notes
1. LOT Polish Airlines v Boeing, US District Court, Western District of Washington (Seattle), trial in May 2026; jury found LOT failed to prove purposeful misrepresentation regarding MCAS disclosure; Boeing cleared of approximately USD 153 million claim. Boeing federal criminal case dismissed November 2025 at DOJ request; Boeing agreed to pay/invest additional USD 1.1 billion in fines, victim family compensation and safety measures. Publicly reported MAX-related DOJ resolutions exceed USD 3.6 billion, comprising the 2021 deferred prosecution agreement of more than USD 2.5 billion and the 2025 non-prosecution/dismissal resolution involving approximately USD 1.1 billion in additional payments, compensation and safety/compliance investment. CNBC, DOJ Criminal Division, NPR reporting. Michael Ryan wrongful death trial scheduled 3 August 2026. Jury verdict of USD 49.5 million to family of Samya Stumo (Ethiopian Airlines crash), May 2026 (Reuters). JD Journal, National Trial Lawyers, Law Fold, Boeing settlement reporting.
2. EASA Notice of Proposed Amendment 2025-07, AI trustworthiness in aviation, published November 2025, original consultation deadline February 2026, extended to March 2026; Level 1 (AI assistance) and Level 2 (human-AI teaming) classification. FAA Roadmap for Artificial Intelligence Safety Assurance. SAE G34/EUROCAE WG114 developing ARP-6983 for AI integration up to DAL-C. Aviation Week, JDA Solutions reporting.
3. Joby Aviation began power-on testing of first FAA-conforming TIA aircraft November 2025 (Joby Q3 2025 results); began flight testing first certification aircraft March 2026 (Reuters). Archer Aviation, Beta Technologies and Wisk Aero also have concurrent FAA certification applications. No US eVTOL manufacturer has received full type certificate for commercial passenger operations as of Q1 2026. Lilium subsidiaries filed for insolvency late October 2024; N.V. board authorised insolvency 4 November 2024; cause was failure to secure EUR 100 million government funding commitment (German parliament budget committee blocked loan); Lilium had received EASA Design Organisation Approval November 2023 (Vertical Mag, AOPA reporting). eVTOL sector has attracted approximately USD 13 billion in investment since 2019 (Reuters, sector reporting). Altitudes Magazine, Airwaysmag reporting.
4. FAA Part 135 certification route used for drone package delivery; operators include Wing Aviation, UPS Flight Forward, Amazon Prime Air and Zipline (FAA). FAA environmental and operational review materials for Zipline projects contemplate up to 400 delivery flights per operating day and, in some cases, 24-hour operations. Proposed Part 108 rule for routine BVLOS operations pending. FAA reporting.