FAA ground stop: Ship AI, and you're the suspect in the next outage

You know the feeling. Three days after you roll out a new system, something somewhere breaks. And before anyone knows the cause, your new system is already sitting in the interrogation room. That is exactly what is happening in the US right now, just at the scale of an entire airspace.

What happened

On Friday the FAA halted all departing flights at US airports. The FAA advisory puts the start at about 12:53 p.m. Pacific. The agency called it a temporary telecommunications outage, after earlier citing communication issues at multiple air route traffic control centers. Several officials told NBC News there was no indication of malicious activity. By late Friday evening Eastern time, the FAA had given no timeline for restarting flights.

The timing is what makes this awkward. On November 10 the FAA said it would use its AI tool SMART across all 22 control centers. SMART suggests routes to cut delays and fuel burn; according to the FAA, an expert team vets the suggestions and controllers make the final call. It had been tested in the Washington, D.C. airspace since September. Politico had reported a 90-day test period before a nationwide rollout, and Gizmodo asked, before the outage, why it was moving faster. The controllers' union NATCA says it was not involved in design or testing.

The denials arrived before the planes moved. Vendor Air Space Intelligence: the ground stop is not based on or a result of SMART. Microsoft, fresh off its own long Microsoft 365 outage: no indication of a link. And Homeland Security Secretary Markwayne Mullin, asked about the AI: you can't rule out anything at this point.

The new AI is guilty until proven innocent

Here is what interests me. Nobody knows today whether SMART had anything to do with it. The story still ran for hours as "Did AI shut down US airspace?" Which means: whoever ships AI becomes the prime suspect for every failure nearby, from day one. Not because the AI did it, but because it is the newest and least understood part of the system.

To me this is not a tech story, it is crisis communications. And it applies to every company wiring AI agents into support, orders or campaigns. AI doesn't fix broken processes, it widens the cracks. In an outage, the first thing it widens is suspicion.

The catch

A denial in the first hour is only worth the evidence behind it. "It wasn't our system" holds only if you can show what the system did that day and what it isn't allowed to do. Roll out faster than planned, leave the people who use it without a say, and you burn exactly that trust. Trust is a brand's most valuable asset, and an outage is where it goes first.

What I would do now

• Write one page for every AI system you run: what it may do, what it may not, who approves. In a crisis, that page is your denial.
• Keep a change log: when each AI feature went live, which update shipped when. Without it, you can't rule anything out either.
• Don't stack an AI rollout on top of other big changes, and stick to the test plan. Cut corners and you own every suspicion.

Bottom lineIf you roll out AI, you need more than a launch plan. You need a plan for the day something else breaks.