Aviation Leaders Draw Line on AI in Flight-Critical Systems

This article summarizes reporting by Aerospace America by Anne Wainscott-Sargent.

Aviation industry leaders convened in San Diego on June 9, 2026, to outline a pragmatic approach to artificial intelligence, emphasizing the deployment of AI for customer service and logistics while intentionally excluding non-deterministic algorithms from flight-critical systems to maintain certifiability.

During a panel discussion at the American Institute of Aeronautics and Astronautics (AIAA) AVIATION Forum, representatives from United Airlines, Reliable Robotics, Collins Aerospace, and the National Aeronautics and Space Administration (NASA) detailed how advanced automation is currently utilized. According to reporting by Aerospace America, the consensus among panelists highlighted a shift away from abstract autonomy concepts toward solving immediate operational friction using classic software engineering for safety-critical applications.

Airline operations and customer management

For major commercial carriers, artificial intelligence is primarily a tool for managing scale and complexity on the ground. Roberta Zimmerman, Director of Air Traffic Strategy, Data Analytics, and Strategic Vision at United Airlines, detailed the operational volume the carrier manages, expecting 5,359 daily departures and offering over 700,000 daily seats across its network. The airline recently achieved a record of 630,500 passengers flown in a single day.

To support this volume, United Airlines utilizes AI to facilitate flight-by-flight customer communication. Zimmerman noted that the technology provides rebooking alternatives for passengers with delayed first legs and calculates predicted walking times between gates at connecting airports. She also cautioned that the national airspace is a complex system of systems, meaning even minor technological integrations require careful management to prevent any loss of operational continuity.

Certification hurdles for flight-critical systems

While airlines focus on passenger logistics, aerospace manufacturers and automation startups face strict regulatory barriers when applying AI to aircraft control. Reliable Robotics, which successfully demonstrated the remote piloting of an 8,000-pound Cessna Caravan from a distance of 50 miles in 2023, is targeting automated operations at approximately 2,000 US airports equipped with Localizer Performance with Vertical Guidance (LPV) capabilities.

Brandon Suarez, Vice President of Unmanned Aircraft Systems (UAS) Integration at Reliable Robotics, explained that using non-deterministic AI in flight-critical systems is currently unworkable for startups due to the lack of established certification standards. Instead, the company relies on traditional software coding languages and classic algorithms for aircraft automation. Suarez described the certification process as the task of convincing an objective expert that a system is correct, a standard that cannot be met if the software’s decision-making process cannot be explicitly explained.

Travis Klopfenstein, Innovation Program Manager at Collins Aerospace, echoed the necessity for explainable systems. He noted that securing funding and leadership approval requires transparent technology. Consequently, Collins Aerospace focuses on increasing automation to optimize human decision-making rather than pursuing full autonomy, while also developing low-criticality applications such as inventory management systems for aircraft galleys.

Establishing reliability standards

The challenge of certifying advanced automation extends to defining acceptable performance metrics. Chester Dolph, an engineer at the NASA Langley Research Center, highlighted that developers must be able to explicitly explain when and why a system works, as well as the specific conditions and reasons for its failure.

Anna Dietrich, an aviation consultant, pointed out the disparity between human and machine performance expectations. She observed that the aviation industry lacks a quantitative consensus on the reliability expected from human operators, who are afforded a margin for error that automated systems are not. Setting the acceptable performance bar for these new systems remains a primary challenge for regulators and developers alike.

AirPro News analysis

We observe a distinct maturation in how the aerospace sector discusses artificial intelligence. The dialogue has moved past the initial hype of fully autonomous passenger aircraft toward a bifurcated reality. On the ground, airlines are rapidly adopting AI to manage the staggering complexity of crew scheduling, irregular operations, and passenger logistics. In the air, manufacturers and startups are deliberately avoiding machine learning in flight control systems to ensure compliance with Federal Aviation Administration (FAA) certification frameworks. Until regulators establish clear, standardized methods for verifying non-deterministic software, we expect the industry will continue to rely on deterministic, classic coding for any system that directly affects safety of flight.

Sources: Aerospace America