Archer Aviation has demonstrated that its ZEE artificial-intelligence platform can predict aircraft movements on airport taxiways and runways several minutes into the future. The company describes the system as an aviation-specific foundation model built on ADS-B data, air-traffic communications, maps, aircraft state, terrain and weather information. Testing is underway at Hawthorne Airport, which Archer controls, and the technology has been shown to commercial partners and regulators. Archer says the goal is to give controllers and pilots more time to react, particularly on the airport surface where runway-related incidents still account for a large share of global accidents.
Prediction is useful. Responsibility is harder. When an algorithm forecasts that an aircraft will hold short or continue onto an active runway, and the forecast is wrong, the chain of accountability is not yet settled. Controllers remain the final authority under current rules, yet they will increasingly operate with machine-generated recommendations that appear authoritative. If a controller follows an incorrect prediction and a conflict develops, the question becomes whether the error belongs to the human, the software provider, the airport that hosted the system, or the data sources that fed it. If a controller overrides a correct prediction and an incident still occurs, the reverse argument appears. Explainability matters because investigators and courts will demand to know why the model produced a particular output. Cyber resilience matters because surface-movement predictions depend on continuous, trustworthy data feeds. Data ownership matters because airlines, airports and air-navigation providers all generate the inputs the model consumes.
Archer frames ZEE as decision support that keeps humans in the loop. That framing is prudent and necessary. It does not by itself answer the liability question. Aviation has decades of experience certifying aircraft systems and procedural tools. It has far less experience certifying predictive models whose outputs shape real-time surface operations. Pilot programmes with government agencies will test accuracy and operational value. They will also force regulators to decide how much weight a prediction may carry, how its performance must be monitored, and who stands behind it when the forecast diverges from reality.
The technology is moving from demonstration to trial. The legal and regulatory architecture that assigns blame when the algorithm misses is still being written.