Lawmakers Urge FAA to Pause AI Air Traffic Management System Over Safety Concerns

WASHINGTON — A group of 25 U.S. lawmakers has urged the Federal Aviation Administration (FAA) to pause its new artificial intelligence-supported air traffic management platform in Washington-area airspace while it answers questions about safety testing, staff training and oversight.

Their October 9 letter to Transportation Secretary Sean Duffy and FAA Administrator Bryan Bedford requests a response by October 19. Signatories include Representatives April McClain Delaney, Don Beyer and Alexandria Ocasio-Cortez, and Senator Chris Van Hollen.

The lawmakers say they have not seen safety evaluation or testing reports and have not been informed of efforts to train aviation safety personnel. They identify the pilot's scope as traffic above Ronald Reagan Washington National Airport (DCA), Washington Dulles International Airport (IAD) and Baltimore/Washington International Thurgood Marshall Airport (BWI).

The FAA began using Strategic Management of Airspace, Routes, and Trajectories (SMART) in a limited mode on September 21. Its launch announcement says FAA staff review every recommendation and local air traffic facility leaders can accept or decline scheduling suggestions. SMART cannot replace controllers or take control of an aircraft, the agency says.

In an email response reported by The Washington Sun on October 13, the FAA said it continues testing SMART, training employees and assessing the program. The agency described specialists sending scheduling recommendations to its Air Traffic Control System Command Center, then routing them to air traffic facilities for local leaders to consider.

What the Lawmakers Want Disclosed

The letter asks whether the FAA made a formal in-service decision before deployment and completed a signed Safety Risk Management Document. It also seeks details of any acquisition or safety-review steps waived or shortened to meet the launch date.

Other questions concern the models SMART uses, how their accuracy was validated, and performance during severe weather, degraded data or system outages. The signatories ask what consultation took place with airport authorities, airlines and the National Air Traffic Controllers Association, and what training personnel received.

The FAA's stated ability for local leaders to decline recommendations leaves a more specific question in the letter: can controllers and traffic management specialists override a suggestion after facility leadership has accepted it? Lawmakers also want to know how those decisions are recorded and evaluated.

The agency says employee training continues, but the materials reviewed for this article do not specify what SMART-specific training local decision-makers completed, how their competence was assessed, or what criteria guide acceptance or rejection of a recommendation. Indeed, we are dealing with a new technological paradigm with many unknowns.

How SMART Fits Into Traffic Management

According to the launch announcement, SMART combines 200 data streams, including weather, flight paths, traffic flow and controller staffing information. Aviation specialists are meant to use its visualizations to anticipate constraints and recommend scheduling changes.

The FAA's SMART fact sheet describes a planning layer alongside existing systems, using airline schedules, airport capacity and airspace conditions to forecast traffic. It updates forecasts and planning recommendations as conditions change. Controllers retain responsibility for keeping aircraft safely separated.

The aviation authority selected Air Space Intelligence in June to supply SMART and Flow Management Data and Services, the planned technological backbone of its Air Traffic Control System Command Center. The company valued the combined award at US$875m over 12 years.

Airways previously examined SMART's role in managing congestion before aircraft leave the gate. The current letter focuses on the evidence and procedures supporting its introduction into operational use.

The signatories describe an initial 90-day evaluation and ask which measures will determine success and justify further deployment. Their request for documents is not tantamount to saying safety testing or training was absent. It seeks the underlying assessments, the rules for human oversight and the benchmarks for expansion.

But human authority alone does not demonstrate effective human oversight. If those reviewing AI-generated recommendations cannot adequately assess the underlying system’s limits and potential failures, their authority to accept or reject its advice offers limited assurance. A pause to establish those limits and test the safeguards is a reasonable precaution.