US ground stop: officials say new AI tool not to blame

US departures were halted for over four hours days after an AI rollout; officials pointed to ageing telecom gear instead.

What happened

At about 3:54pm Eastern time on Friday, the US Federal Aviation Administration stopped departing flights across the entire country. It was the first nationwide ground stop since the NOTAM system failure of January 2023. Transportation Secretary Sean Duffy said more than 3,500 aircraft were in the air when the order went out. By 7:05pm at least 827 flights had been diverted, according to Cirium data cited by the Associated Press.

Suspicion quickly turned to artificial intelligence. Earlier that week, the FAA had said it had expanded a new AI-powered tool, SMART (Strategic Management of Airspace, Routes and Trajectories), to all 22 of its en-route air traffic control centres. SMART, built by the contractor Air Space Intelligence, analyses airline schedules, weather and airport capacity to predict congestion and suggest fixes. It does not replace controllers, and people decide whether to use its recommendations. Futurism reported that the system cost the government just under $1 billion.

Asked about the timing at a news conference, Homeland Security Secretary Markwayne Mullin said, "We're looking at that." Air Space Intelligence said the ground stop was not caused by SMART or its integration with the FAA.

By the evening, officials had a different explanation. AVweb reported that they blamed overloaded switches and routers in the FAA's ageing telecommunications equipment, while newer digital systems kept working. Duffy said SMART only gives recommendations to controllers and is not connected to the FAA's STARS or ERAM air traffic systems, and that there was "no possible way" it had played a part. Mullin said investigators had found no evidence of a cyberattack or AI-related interference, though the investigation is still going on and the root cause has not been confirmed. Duffy said the FAA never fully lost radar or communications, and medical, critical cargo and law-enforcement flights were exempted. The ground stop was lifted at 8:15pm, more than four hours after it began, and airlines warned of knock-on delays.

Why it matters

This is a story about where failures really come from. The newest, most visible system got the blame first. Officials' early explanation points to old equipment that nobody talks about until it breaks, although the root cause is still being investigated.

It is also a sign of how nervous people are about AI in critical systems. Within hours, an AI tool that only makes suggestions was being linked to a nationwide shutdown. Any company that rolls out AI should expect the same reflex from customers, staff and the press the next time something goes wrong.

What it means for your business

Two lessons apply well beyond aviation. First, be able to say clearly what your AI can and cannot touch. Officials could rule SMART out quickly because it was advisory and separate from the systems that failed. If your AI tools have broad access to everything, you will not be able to give that answer, and you may not know it yourself.

Second, do not let AI projects take all the attention and budget from the systems underneath them. Networks, servers and old software still carry the load. For manufacturing, logistics and other operations-heavy firms adding AI on top of legacy systems, the boring maintenance work is part of the AI plan, not separate from it.

What you can stop worrying about: so far there is no evidence that an advisory AI tool caused this failure. Keeping AI advisory and separate from core systems is what made it quick to rule out.

What to do this week

Write a one-page map of each AI tool you use: what data it reads, what it can change, and who approves its output. Ask your IT lead which core systems are oldest and least maintained, and what happens if one fails. Prepare a short holding statement for customers and staff for the day something breaks and AI gets the blame.