One word. Hallucination.
That single term, delivered with the gravity of someone who has spent decades guiding planes through congested airspace, captures the unease now rippling through air traffic control circles. As the Federal Aviation Administration rolled out its new artificial-intelligence tool this week, a former controller with experience from the chaos of 9/11 voiced a stark warning. The system, he fears, could invent problems that don’t exist or miss real ones, with consequences too severe to ignore.
The tool in question is SMART. Short for Strategic Management of Airspace, Routes and Trajectories, it went live Monday at the three major airports serving the Washington area: Ronald Reagan Washington National, Dulles International and Baltimore/Washington International. Politico reported that Transportation Secretary Sean Duffy and FAA Administrator Bryan Bedford unveiled the system with assurances it would sharpen decision-making without displacing human controllers. “Humans control the airspace, humans make decisions,” Duffy said during the launch event. The technology, part of an $875 million contract with Air Space Intelligence, analyzes schedules, weather, airport capacity and other factors to spot potential conflicts days or weeks ahead.
But not everyone shares the optimism. S. Todd Yeary, who directed traffic on 9/11 and spent years in the role, told NewsNation he feels “very conservative” about trusting the system so soon. His primary dread centers on the well-documented tendency of AI models to hallucinate. “One word: hallucination,” he said, echoing the concerns detailed in Futurism’s coverage of his interview.
Yeary’s point lands with force. Air traffic control operates in a fluid environment where data shifts constantly. A pilot requesting visual flight rules service might not appear in airline schedules. Weather can change abruptly. If the AI draws on incomplete or outdated inputs, it could generate recommendations based on phantom threats. Or worse, overlook genuine risks. “There has to be some room for there to be an override,” Yeary insisted. Controllers must retain the power to veto suggestions. Without that safeguard, confidence erodes fast.
The rollout arrives against a troubled backdrop. U.S. air traffic control has faced persistent staffing shortages, aging equipment and a string of high-profile incidents. A January 2025 midair collision over the Potomac that killed 67 people drew fresh scrutiny to FAA practices in the busy D.C. corridor, The New York Times noted. Outages at facilities handling Newark traffic exposed fragile radar and communication systems. A Government Accountability Office report released this week warned that aircraft communications remain “incredibly insecure,” vulnerable to spoofing and jamming that could send fake clearances or disrupt operations, according to Reuters.
And yet the pressure to modernize is intense. Delays and cancellations cost airlines, passengers and the economy billions annually. Duffy has promoted SMART as a way to predict bottlenecks 45 days out, shifting flights by minutes to prevent ground holds. Bedford described it as giving controllers “much better decision-making capabilities.” The system centralizes some 200 data streams into a shared view. In theory, everyone from FAA facilities to carriers operates from the same picture.
But theory collides with practice when AI enters the equation. Large language models and predictive systems can produce confident-sounding errors. A recent military case, reported by TechCrunch, showed an AI chatbot generating false intelligence about a Chinese vessel that nearly triggered an armed response before officials caught the mistake. The parallel is uncomfortable. In aviation, the margin for such errors shrinks to seconds.
Industry insiders expressed private alarm before the launch. Politico’s earlier reporting revealed airlines “panicking” over initial plans that seemed to grant the AI broader authority to adjust schedules and routes. The final version appears more modest. It offers recommendations that local controllers can accept or ignore. A 90-day test period in the D.C. region will gauge performance before wider expansion. Still, Rep. Don Beyer, a Virginia Democrat whose district includes Reagan National, called for suspension. He argued controllers received inadequate consultation and training. “The FAA should not use my constituents as guinea pigs for an unproven AI air traffic control system,” he said, as covered in Newsweek.
Proponents counter that human oversight remains absolute. The tool does not issue commands to aircraft or replace controllers. It functions as an advanced advisor, scanning horizons controllers cannot see amid the press of immediate duties. Early data from similar predictive systems in other industries suggest gains in efficiency. Fuel savings, fewer overtime hours, smoother recovery from storms. Those benefits could translate to lower fares over time.
Yet the former controller’s caution refuses to fade. Yeary stressed that real-world aviation includes variables no schedule captures. VFR traffic. Sudden pilot requests. Equipment quirks. He questioned the data foundation itself. Where does the system pull its information? How does it handle gaps or contradictions? “Air traffic control is a constantly changing environment,” he observed. Assumptions of static reliability invite trouble.
This tension reflects a larger shift. Aviation has long relied on rigid procedures, redundant systems and highly trained professionals. Introducing probabilistic AI demands new thinking about trust, verification and accountability. What happens when a recommendation looks plausible but stems from a hallucinated pattern? Who bears responsibility if controllers follow it and something goes wrong?
The FAA insists safety stays paramount. Bedford and Duffy repeated the message: the technology augments, it does not supplant. Plans call for national deployment by 2028, paired with hardware upgrades and efforts to hire more controllers. Training failure rates remain high. Staffing shortfalls persist at many facilities. The AI arrives as one piece of a broader fix for a system strained by growth and complexity.
Even so, the timing feels uneasy. Monday’s launch coincided with ground stops and communication glitches along the East Coast, though those stemmed from a severed cable rather than the new software, Futurism reported. The coincidence amplified skepticism on social media and among pilots. One controller union representative described the rollout as moving too quickly without sufficient input from those who will use it daily.
Research into AI for air traffic management offers mixed signals. Academic work on models like ChatATC, discussed in Aerospace America, shows promise for analyzing historical delay programs. But those same systems can generate incorrect answers. Some researchers even encourage controlled hallucination to simulate rare disruption scenarios for training. The distinction between helpful creativity and dangerous invention blurs in live operations.
For now, the experiment unfolds in one of the nation’s busiest airspaces. Washington airspace already carries heavy military, commercial and general aviation traffic. The 2025 collision underscored vulnerabilities. If SMART delivers on predictions and eases pressure on overworked controllers, it could mark a genuine advance. If hallucinations surface, or overrides become routine because trust is low, the backlash could slow modernization for years.
Yeary’s message carries the weight of experience. He doesn’t oppose technology. He questions blind faith in it. “Controllers must be able to step in,” he said. That requirement feels non-negotiable. The skies remain too unforgiving for anything less.
Additional recent reporting from The Washington Post highlights the tool’s goal of spotting congestion earlier through real-time data synthesis. Whether that synthesis proves reliable under pressure will determine if this week’s debut represents progress or a cautionary tale.