FAA Traffic AI at Three Washington Airports Faces Safety Audit Request

A Federal Aviation Administration traffic-planning system operating around three major Washington-area airports now faces a formal request for independent scrutiny of its testing, reliability, controller training and safety safeguards. On October 13, 2026, 23 Democratic members of the House Transportation and Infrastructure Committee asked the Government Accountability Office to audit the FAA’s Strategic Management of Airspace, Routes and Trajectories system, known as SMART, and the related Flow Management Data and Services platform.

The request is not a finding that either system is unsafe, and it does not establish that the GAO has opened an audit. The FAA has not suspended SMART, flights have not stopped, and no special flight restrictions have been imposed because of the system. SMART began operating September 21 in airspace serving Ronald Reagan Washington National, Washington Dulles International and Baltimore/Washington International Thurgood Marshall airports.

What SMART does and does not do

SMART analyzes airline schedules, weather, airport capacity and restrictions elsewhere in the national airspace system. Its predictive models identify potential congestion or disruption and recommend changes to routes or flight timing. The associated FMDS platform is intended to modernize the underlying flow-management environment in which those recommendations are used.

This is strategic traffic management, not autonomous aircraft control. The FAA says SMART does not separate aircraft, issue instructions directly to pilots or replace controllers. Human operators retain authority over operational decisions. That boundary is important, but it does not settle the central assurance question: whether people can safely judge automated recommendations under real operating pressure.

Human supervision is a safeguard only when operators understand what a system is designed to do, where its performance limits lie and how to respond when its output conflicts with other information. A recommendation can influence a decision without becoming a formal aircraft-control command. That makes training, interface design, alert presentation and confidence calibration part of the safety case rather than secondary implementation details.

What lawmakers want independently examined

The requested review would examine the FAA’s safety-risk assessments, testing and certification procedures, reliability criteria, deployment strategy and consultation with controllers, airlines and other stakeholders. The committee members also asked about controller training, performance during severe weather or outages, possible waivers of safety-review requirements and the standards governing an initial 90-day evaluation.

Those categories address different layers of system assurance. Testing asks whether the software performs as intended under defined conditions. Reliability criteria establish how consistently it must perform. Training determines whether users recognize appropriate and inappropriate recommendations. Deployment strategy governs how quickly operational exposure expands before enough performance data are available to evaluate assumptions made during development.

Severe weather and outages are particularly relevant because traffic-management tools deliver their greatest potential value when the network is disrupted and when operating conditions are least routine. A useful evaluation therefore must distinguish normal-day accuracy from performance during rapidly changing constraints. It also needs measurable thresholds, not simply demonstrations that the software can generate plausible options.

Lawmakers have raised the possibility that inaccurate AI-generated information could interfere with safety-sensitive decisions. That remains an attributed concern, not a documented defect. No SMART-related accident or operational failure has been reported here, and there is no official finding that the system is unsafe.

The tradeoff behind the Washington deployment

Deploying SMART in busy airspace can provide a demanding and information-rich test environment. It also increases the importance of clear operating limits because recommendations may affect airlines, traffic managers and controllers across three interconnected airports. The engineering policy question is not simply whether testing should occur in complex airspace, but what evidence and safeguards should be required before and during that exposure.

The FAA awarded a 12-year, $875 million contract covering SMART and FMDS. The intended operational benefits include better routing, reduced fuel use, improved on-time performance and faster recovery from weather or congestion. If the tools work as intended, passengers and airlines could benefit from fewer network disruptions. If recommendations are inconsistent or poorly understood, operators may face additional workload while checking or rejecting them. The balance depends on measured performance and human-factors results that have not yet been independently resolved by the requested review.

FAA spokesperson Hannah Walden said on October 14 that lawmakers have a standing invitation to observe a demonstration. A demonstration can clarify system functions and the division of authority, but it is not a substitute for examining test records, training standards, reliability metrics and performance under abnormal conditions.

SMART therefore remains in an unusual but clearly defined position: it is active decision-support software, not an autonomous controller, while its evidentiary foundation is being challenged through the oversight process. The next consequential step is whether the GAO accepts the request and independently tests whether the FAA’s validation, training and deployment controls match the demands of the airspace where SMART is already operating.

By Thomas Caldwell — AMI’s senior editor for mechanical and mobility engineering, covering vehicle electronics, systems integration, electrification, chassis systems, propulsion, and safety policy.