Blog · 19 min read ·
ShareFAA and Microsoft Outages: It Wasn't AI. It Was a Change.
Update — November 16, 2026: Microsoft has now given a cause, and the FAA's has narrowed without being settled. Microsoft published a preliminary post-incident report for MO1550821 on November 15, and we have read it in our own tenant. It traces the outage to a latent defect in a fix for Microsoft 365 network gateway servers, deployed from July 2026, that a routine Teams failover exposed while nearly 14% of gateway capacity was out of service. The failure cascaded across the fleet and could not be quickly rolled back. Microsoft's final report is due within five business days. At the FAA, the agency has now published its own account: data flooded its network from two sites, a radar site in Amarillo, Texas and a radio site in El Dorado, Arkansas, and it switched off a broadcasting function on the new routers. On Monday, Transportation Secretary Sean Duffy called the Arkansas router the "likely cause" while FAA staff and FBI agents examined it at L3Harris's lab in Florida. No root cause has been announced. Details are in the sections below, and a full source list is at the end.
Two large outages landed in the same week. On Thursday, November 12, Microsoft 365 went down. On Friday, November 13, the FAA stopped departures nationwide. Both organizations were asked whether AI or a cyberattack was involved, and both said no on the record. What they described instead is more ordinary, and more useful to you: someone changed something on a network, and the failure traveled over a path that everything depended on.
The two outages were not connected to each other. As GeekWire reported, Homeland Security Secretary Markwayne Mullin said there was "no connection between this and Microsoft," and Microsoft posted on X that it had "no indication that the FAA's nationwide ground stop is related to any Microsoft services." We put them side by side because they failed in a similar way, not because one caused the other.
What happened at the FAA
According to Yahoo News' live coverage, Transportation Secretary Sean Duffy said the outage began while L3Harris Technologies, which runs the network, was switching lines from copper wire to fiber at a remote FAA site, and that a router involved in the work flooded the rest of the system, disrupting communications with airplanes and radar. Weather.com separately tied the communications outage to "an update on an L3Harris-operated network used by air traffic control." L3Harris told Yahoo News it is "supporting the FAA in its investigation," and we are reporting its role only as officials described it.
The FAA halted departures nationwide for more than four hours, which Yahoo News called the longest ground stop since the September 11, 2001, attacks. About 12,000 flights could not be completed, FAA Administrator Bryan Bedford said. The FAA's own questions-and-answers page puts the start at "approximately 3:30 p.m. ET" and says air traffic resumed "as of 8:15 p.m." It says operators canceled "approximately 12,462 flights" that Friday, and that "not all cancellations were directly caused by the nationwide ground stop."
Officials are still working out whether the cause was human error or equipment. On Saturday, Duffy said they are checking "whether we had incompetence," meaning human error by the contractor, or a faulty router: "We haven't determined what the fault was." Bedford said the review will determine whether the router was installed or configured incorrectly, or was simply defective.
What has narrowed since November 14
By Sunday the reporting was more specific, though still not a conclusion. According to Fox News, Duffy said on "Fox News Sunday" that two routers, from FAA facilities in Texas and Arkansas, are undergoing extensive testing, and that investigators have found no indication of a foreign cyberattack. "We think it was the installation or the router that was the bad actor," Duffy said. "But we're open, if we see something else, we're gonna continue to investigate with our federal partners and law enforcement." Fox News reported that L3Harris secured the two routers and is running diagnostic testing at its headquarters in Florida, with engineers from the FAA, Juniper and Cisco assisting. According to Yahoo News, Duffy said on ABC's "This Week" that when the new router in El Dorado, Arkansas, was turned on, a router installed about three weeks earlier in Amarillo, Texas, "started to flood our system as well."
On the mechanism, Yahoo News reports: "Officials believe a feature that broadcast data more widely than necessary caused the overload. Duffy said the FAA has turned that feature off on all 69 routers of that type in the network." Other outlets put that count at roughly 60.
The FAA's own account
The FAA has published a questions-and-answers page, dated Sunday, November 15. We read it on faa.gov, and text in quotation marks in this section is the FAA's. It says: "The FAA's legacy equipment became unstable when data began flooding the FAA FTI network, at a rate of 1,000 connection attempts every 4 to 5 seconds." The initial investigation points to "two potential locations": "a radar site in Amarillo, Texas and a radio communications site in El Dorado, Arkansas," where L3Harris "had been doing work … including deployment of a new router." When the data went wrong, "L3 Harris blocked those two locations and the network stabilized. The broadcasting function was also disabled on all the new routers as a precaution." The FAA says it disabled "the same functionality on approximately 60 similar routers," and that instability began at "17 of our 22 Air Route Traffic Control Centers."
Two answers on that page bear on the lessons below. Asked whether its upgrades caused the outage, the FAA says: "No, this is due to the legacy TDM system causing instabilities." That sits alongside officials' account that the flooding began when a new router was switched on; we report both and do not reconcile them. Asked why the router work was done in the middle of the day, the FAA says the replacement "has regularly been done for decades, without any impacts," and that "based on what happened Friday, we will reevaluate when we do replacement/maintenance work." On the cause, the FAA says it and L3Harris "continue to investigate the root cause and ensure this doesn't happen again."
Monday, November 16
Duffy posted on X that "Our FAA team is currently on site at the L3Harris lab in Florida analyzing the router that was the likely cause of Friday's nationwide ground stop," as WINK News reported. The Next Web reports that FBI agents are also there, to rule out a cyberattack, according to Duffy. Hewlett Packard Enterprise, which owns Juniper, defended the device; a spokesperson told ABC News, as The Next Web relays: "Based on what we know today, the device functioned as it was configured to by the installer. It did not fail or act improperly." As of Monday afternoon, Yahoo News' live coverage still described the root cause as unknown.
Accounts of the equipment differ. Sources told CBS News on November 14 that the FAA believed the problem was limited to Cisco 600 series routers made between 2001 and 2006, that they were receiving conflicting information about the cause, and that an update to the FAA's private communications network caused a reboot and a crash. Later, Yahoo News relayed that, according to The Wall Street Journal, the Arkansas router was a new Juniper Networks device installed by L3Harris, and that Juniper's owner, Hewlett Packard Enterprise, told the Journal: "They did not fail." We did not fetch the Journal's report.
Whether the fault was installation or equipment is still open. The New York Times reported on Sunday that investigators had not determined whether a faulty part or an error by L3Harris caused the initial problem in Arkansas, and that Duffy said they were equally unsure why the Amarillo facility failed when no upgrade work was being done there at the time. When asked, Duffy could not guarantee that a similar failure would not happen again, the Times reported. L3Harris has said it is supporting the FAA's investigation. The Dallas Morning News adds, as background, that L3Harris was partially blamed for a September 2025 air traffic control outage in Dallas. That is context, not a finding about this one. We are not taking a side.
What happened at Microsoft
Microsoft 365 incident MO1550821 (Exchange Online: EX1550543) knocked out Outlook and a long list of services that depend on Exchange Online. We covered the details in our Microsoft 365 outage post, so here is only what matters for this argument.
On Microsoft's public status page, an admin-center post at 17:57 UTC on November 13 said: "This was a technical issue not related to cyber or AI / agents. The incident was introduced by an issue during an earlier network capacity upgrade that surfaced following a recent network change." The same text appeared on the consumer status page at 18:57 UTC. Earlier, at 19:30 UTC on November 12, Microsoft said it had "paused all deployments and configuration changes." At 01:06 UTC on November 13 it said it had "begun reverting a recent network configuration change that correlates with the start of impact."
Microsoft's final post on the consumer status surface, last updated at 03:51 UTC on November 14, says: "Service restored." It gives a start time of Thursday, November 12, at 10:05 AM UTC and an end time of Friday, November 13, at 4:10 PM UTC, and calls itself "the final update for the event." That consumer post carries no incident ID; the business incident had no final public post as of November 14. BleepingComputer reported a start of 1:26 PM ET, which does not match that 10:05 AM UTC figure, so we are not stating a single duration. Microsoft's preliminary post-incident report followed on November 15. It gives a fuller account than the status text, and its own window: 4:20 PM UTC on November 12 to 4:10 PM UTC on November 13.
What Microsoft's preliminary report says
Microsoft's preliminary post-incident report is dated November 15 and covers both incident IDs, EX1550543 and MO1550821. It sits in each affected tenant's admin center under Service health, so it cannot be linked; we read it in our own tenant, and text in quotation marks is Microsoft's. The final report is due within five business days of resolution, with the full list of repair items dated November 20.
Microsoft confirmed the release publicly. Its Microsoft 365 Status account posted on X on November 16: "We have released the Preliminary Post-Incident Report of last week's outage, which is accessible via the Admin center under MO1550821 and EX1550543." Charles Lamanna, a Microsoft executive vice president, posted an abridged version and wrote: "We are deeply sorry for the impact to all our customers and recognize the impact that this has on their normal business & operations." If you administer a Microsoft 365 tenant, the full report is under Service health, in the incident history for either ID.
Microsoft names three factors. The first is a latent defect: "lock contention inside an internal Windows process which caused the machine's resources (CPU & memory) to be exhausted." It came from "a fix that was prepared specifically for Microsoft 365 networking gateway servers; neither the underlying defect nor the fix applies to broader Windows deployments." It began deploying in July and ran worldwide for more than three months without impact. Testing missed it: "The contention issue is worsened on machines with more cores and newer hardware; whereas our load testing tended to focus on weaker hardware with fewer cores."
The second and third are ordinary on their own. At 01:07 UTC on November 12 (5:07 PM PT, November 11), an automated failover moved Teams traffic onto the gateway shared with Exchange Online, which Microsoft calls "part of normal operations." And "nearly 14% of server capacity for the networking gateway was out of service due to an unrelated issue from the previous day." Microsoft says "The service had sufficient capacity headroom, so the reduction in server capacity had no impact on its own." Together they hit the defect during the US traffic peak.
Then the safety mechanism made it worse. Traffic management moved users off failing servers onto healthy ones, which "increased the likelihood for the same problem to occur on the remaining endpoints." In Microsoft's words: "In less than two hours, the previously unknown defect was encountered across our fleet and rendered them unresponsive. At its lowest, more than 80% of servers were unresponsive." The bulkheads that normally separate capacity units and regions "did not stop the cascading failure," due to a separate issue.
Recovery was hard. "Our normally automated recovery processes were not designed to restore service from this broad unresponsive state." Because the defect was months old, "our normal rollback, flighting, and feature flags were unable to quickly revert the change." Engineers built "an update that returned the code path to the previously known good state in Windows," and applied it to large groups of servers at a time. Teams, OneDrive, SharePoint and Office apps failed out of the gateway and recovered earlier, but "not all services, like Exchange Online, can fail out of this network gateway for scale reasons." The report's timeline has the connectivity monitor flagging an increase in issues at 16:03 UTC on November 12 (8:03 AM PT) and the mitigation, "an update that reverted the defective change," first applied at 07:20 UTC on November 13 (11:20 PM PT, November 12).
Our reading, which is ours and not Microsoft's: this is the thesis of this post in Microsoft's own words. A change passed testing, sat latent for months, and could not be quickly rolled back when it failed, and a failover designed for resilience spread it. It is also a different account from the status-page text above, which pointed to "a recent network change." The two may describe the same chain of events, and the final report may clarify. We are not reconciling them.
Was it AI? Both said no
The question came up because the FAA had just begun rolling out an AI traffic-management tool. Scripps News quoted Bedford: "We see zero evidence of anything nefarious and zero evidence of anything AI-related." AOL reported that the FAA had begun rolling out an AI tool to monitor air traffic and prevent delays. Duffy said there was "no possible way" it could have affected the outage, and the tool's vendor denied any connection.
Microsoft's wording, quoted above, is just as direct: "not related to cyber or AI / agents." We take both statements as the organizations' own accounts. In these two cases, the people closest to the evidence pointed at a network change, and that is where a mid-market IT team can act.
What the two have in common: a change, one shared path, slow diagnosis
- A change started it. At the FAA, line work from copper to fiber, per Duffy. At Microsoft, the status page pointed to a network capacity upgrade and a recent network change. The preliminary report traces the failure to a fix deployed from July that had been in production for months. At the FAA, Duffy said a router installed about three weeks earlier in Amarillo also flooded the system once the new Arkansas router was turned on.
- One shared path carried the damage. At the FAA, officials say a router flooded the rest of the system. Bedford said all facilities rely on one network, and that "there's no backup to the backup." At Microsoft, Exchange Online sat under Outlook, Teams calendars, Copilot, and security tooling, according to Microsoft's own impacted list.
- Finding the cause took time. The FAA's maintenance is logged on paper, so officials had to search through paper records to figure out who was working where. Tracing the problem took about 45 minutes. Microsoft told customers on November 12 that it had paused deployments and configuration changes, then spent hours before reporting it had begun reverting a change. The preliminary report puts the first connectivity alert at 16:03 UTC on November 12 and the first applied mitigation at 07:20 UTC on November 13.
Change control, redundancy, and logging are hard even for well-resourced operators. A 150-seat firm with one firewall and one ISP is exposed in the same shape at a smaller scale.
The 2009 precedent
This has happened before. A June 2010 letter from the Department of Transportation's Office of Inspector General describes November 19, 2009, when "129 of the Nation's air traffic facilities experienced an outage of the Federal Aviation Administration's (FAA) telecommunications that delayed thousands of travelers and grounded hundreds of flights nationwide." In that case "a Harris technician incorrectly configured an FTI router" at Los Angeles Center. Harris Corp. is L3Harris's predecessor.
Two lines in that letter read like a checklist for today. "Service restoration was delayed for 5 hours because an automatic tool that alerts technicians to network failures and their locations did not work as intended." And: "FAA's oversight of the FTI contractor could have been more effective. FAA was unaware that Harris officials had configured the network in error and made other procedural errors." The 2026 review is open, so we are not saying the causes match. The pattern is familiar: contractor-performed network work, a router, and weak visibility into what was changed.
1. Put change control around ISP, firewall, and WAN work
Change management does not need a heavy process. For anything that can take a site or a service offline, such as an ISP cutover, a firewall replacement, a WAN or SD-WAN change, or a DNS or mail-routing edit, require five things:
- A change window. Asked why the work was done on a Friday afternoon, Duffy said work at small, remote sites is done during the day because it is not expected to affect the wider system, and that "there'll be a reevaluation of when we do this work." The FAA's own page says the same: the replacement work "has regularly been done for decades, without any impacts," and it "will reevaluate when we do replacement/maintenance work." Your lesson: decide the window by what could go wrong, not by how small the site looks or how many times the job has gone fine.
- A rollback plan. Before the work starts, write down the exact steps to return to the previous state, and who decides to pull the trigger. Microsoft's own response included reverting a recent network configuration change. And a rollback plan has a shelf life: in Microsoft's preliminary report, because the defect was months old, its "normal rollback, flighting, and feature flags were unable to quickly revert the change." Ask what you would do if the change you need to undo is not the latest one.
- Who is on site, and who is on call. Name them. Include the vendor's technician and a way to reach your own staff while the work is under way.
- A written log you can search. Tracing the FAA's problem took about 45 minutes, which Bedford said was partly because maintenance is logged on paper. Use a ticket or change record with the date, system, person, and company for every change, and make sure someone on your side can find it in two minutes.
- The same rules for vendors and contractors. Work done by an outside provider is still your outage. Ask that notice of planned work reaches you first, and read who owns what in your MSP contract before you need to.
2. Map every single point of failure
A single point of failure is easier to find on paper than during an outage. List the components below, then write next to each one what happens to the business if it disappears for four hours and for a day.
- One ISP circuit. A second circuit from a different provider helps only if it follows a different path into the building.
- One firewall. A spare in a closet is a plan only if someone has loaded its configuration.
- One identity provider. If sign-in to Microsoft 365 fails, so does everything that trusts it. Keep emergency access accounts and know how they work. AWS had a version of this failure on November 15 (see below).
- One mail path. The November 12 incident showed how much leans on Exchange Online. Our September 30 Azure gateway post covers a related shape: a VPN backup that shared the failure of the thing it backed up.
- One remote-management tool. Your provider's remote monitoring and management platform can reach every endpoint, which makes it a high-value target and a dependency. See questions to ask your provider about RMM.
Update — November 15, 2026: a third outage, and this one hit sign-in. Two days after the FAA ground stop, AWS reported on its Health Dashboard that “Between 3:05 AM and 6:10 AM PST we experienced signup errors, authentication issues, and connectivity issues in the US-EAST-1 Region. Customers using IAM Identity Center for authentication into the AWS Console, or for authentication into other AWS Services … would have been affected.” For sign-in, AWS wrote during the incident that “There are no workarounds for impacted customers.” AWS says it fixed sign-in at 5:20 AM PST. A load-balancer issue in one availability zone ran until 6:21 AM PST, later than the 6:10 AM in AWS's own summary. AWS has not published a cause, so we are not calling this a change. The lesson is the one in the identity bullet above: when the sign-in service itself fails, every system behind it fails with it, and nothing on your side fixes it. For a Microsoft 365 firm that service is Entra ID. Emergency access accounts cover a different case, a lockout on your side from Conditional Access, MFA or federation, and you need them for that. For an outage at the identity provider itself, the defense is knowing which systems keep working without it and having a manual or offline fallback written down for the ones that matter.
Redundancy and recovery are different jobs. A second copy in the same region or tenant does not protect you from the wrong change. Our post on why redundancy is not backup goes deeper, and restore testing is how you learn whether the fallback works before the day you need it.
3. Write a business continuity plan for "the cloud is up but email isn't"
Most continuity plans assume a building is lost or a server is down. November 12 was a different case: for many customers the internet worked and Microsoft was running, but mail and the services that lean on it were not. Microsoft's own status text said service access can recover before mail flow fully normalizes. A business continuity plan for this case is short:
- How staff reach each other if email is out (phone tree, a chat tool that does not depend on the same service).
- Where customer, vendor, and after-hours contact lists live outside the mailbox.
- Who decides to switch to the fallback, and who tells staff and customers.
- What "recovered" means, so nobody declares success while queued mail is still arriving.
- A rule against working around the outage with personal accounts. Data sent that way does not come back.
Where AI does fit: it speeds up changes
We found no evidence in either account that AI caused these outages. AI is still relevant to this topic because it makes changes faster, and faster changes need the same controls. As reported by Computerworld, an Amazon senior vice president "acknowledged that 'GenAI tools supplement or accelerate production change instructions, leading to unsafe practices,'" and Computerworld reported that Amazon's AI safeguards "are not yet fully established." Computerworld also reported that an AI agent's fix to "delete and recreate the environment" was said to have triggered "a roughly 13-hour outage that hit the AWS Cost Explorer service in parts of mainland China." Amazon has insisted the root cause was "user error" and misconfigured access controls. We are relaying a reported dispute about cause, not settling it.
The practical rule is plain: treat an AI-assisted change, or an AI agent with permission to make one, like any other change. It gets a request, an approver, a window, a rollback, and a log entry. It does not get broader access because it works quickly. Our AI acceptable use policy guide shows how to tie each rule to a control, and the Claude outage post covers the other side: what happens when an AI service sits in the critical path of your own work.
Where we fit
We don't build networks for the FAA, and we can't fix Microsoft's. For regulated mid-market organizations, we do the work around those failures: write the change-control procedure that covers your vendors, map your single points of failure, test restores, and draft the continuity plan that says who does what when email is down. Vendor-neutral and labor-only.
Sources
We link only pages we fetched and read. FAA statements quoted from its questions-and-answers page are the FAA's own words; other FAA and DOT statements are as quoted by the outlets listed here.
Primary: the FAA
- FAA: Nationwide Ground Stop: Questions and Answers (dated November 15, 2026; read November 16)
Officials and the contractor (via reporting)
- WINK News: FAA probing router connected to Florida based contractor L3Harris after Friday's nationwide ground stop (November 16, 2026; includes Duffy's post on X and L3Harris's full statement)
- The Next Web: FAA ground stop cost ‘billions’, Duffy says, as router maker defends device (November 16, 2026)
- GeekWire: Microsoft declares outage over as the world digs out of its inbox (the “no connection” statements from DHS and Microsoft)
- Fox News: Duffy: Investigators examining 2 routers after nationwide FAA ground stop, no signs of foreign cyberattack (November 15, 2026)
- CBS News: Ground stop shows "we have to get off old network," Transportation Secretary Sean Duffy says (November 15, 2026)
- The New York Times: Duffy Points to Router in Arkansas as Start of Nationwide Ground Stop (November 15, 2026)
- CNBC: ‘Root cause’ of nationwide ground stop is unknown, FAA says (November 14, 2026; includes L3Harris's statement)
- The Register: FAA blames ground stop on faulty copper wiring (November 13, 2026, updated; includes L3Harris's statement)
- The Dallas Morning News: report on the L3Harris router in Texas and the FAA's statement (quotes the FAA's website statement)
- NPR: Duffy says work by a contractor to upgrade old copper wires led to the groundstop (November 15, 2026)
Investigation and cause
- Yahoo News: FAA ground stop latest updates: Flight cancellations ease; Duffy seeks $50 billion for air traffic control overhaul; root cause still unknown (live blog, read about 7:20 PM UTC, November 16, 2026; relays The Wall Street Journal, which we did not fetch)
- The Verge: America's air traffic control system couldn't handle a single bad router (November 16, 2026)
- The New York Times: How Could a Router Lead to a Nationwide Ground Stop? (November 15, 2026)
- NBC News: How a router problem at one FAA site led to a nationwide ground stop (November 14, 2026, updated November 15)
- USA Today: FAA contractor, router focus of investigation into air traffic halt (November 14, 2026, updated November 15)
- CNBC: Duffy ties router in southern Arkansas to nationwide ground stop (November 14, 2026)
- CBS News: FAA lifts nationwide ground stop that disrupted thousands of flights; White House says no indication of "nefarious act" (November 14, 2026)
- Fox Business: Monday air travel faces key test after FAA ground stop as possible cause emerges (November 16, 2026)
- Spiceworks: The FAA ground stop started with a problem IT pros recognize all too well (November 16, 2026)
- CNN: What we know about the cause of the nationwide FAA ground stop (November 14, 2026)
- ABC News: FAA ground stop live updates: Flight cancellations continue into Monday (updated November 15, 2026)
Context
- Scripps News: What caused the FAA ground stop? Officials point to router failure during update
- AOL: Both FAA and Microsoft rolled out AI updates before outages (November 14, 2026)
- Weather.com: nationwide ground stop coverage (live page; headline updated since November 13)
- NPR: FAA blames Friday's ground stop on ‘equipment issue’ during telecoms network upgrade (November 14, 2026)
- CBS News: Airline schedules returning to normal after nationwide ground stop, FAA says. Here's what to know. (updated November 16, 2026)
- BBC: Copper wire upgrade behind US flight suspension, transport secretary says
- BBC: How telecoms upgrade work sparked US flight chaos
- Business Insider: Here's What Officials Say Caused the FAA's Nationwide Ground Stop
- Newsweek: profile of L3Harris and its FAA contract
- CNN: Duffy blames prior administrations and dated systems for airspace meltdown (live coverage, November 15, 2026)
- DOT Office of Inspector General letter CC-2010-012, June 17, 2010 (the November 19, 2009 FTI outage)
Microsoft, AWS and Amazon
- The Register: Microsoft blames monster email outage on software defect introduced in July fix (November 16, 2026; first public report on the preliminary post-incident report)
- Microsoft, Preliminary Post Incident Report for Microsoft 365, EX1550543/MO1550821, report date November 15, 2026 (Microsoft 365 admin center, Service health; requires tenant sign-in, so not linkable)
- Microsoft 365 Status on X (November 16, 2026; announces the preliminary report)
- Charles Lamanna on X (November 16, 2026; abridged summary and apology)
- Petri: Microsoft Explains Causes of Microsoft 365 Outage (November 16, 2026)
- BleepingComputer: Massive Microsoft 365 outage disrupts Outlook, Teams, Copilot worldwide (reported start time)
- Microsoft's public status page, status.cloud.microsoft (posts captured November 12 through 14, 2026)
- AWS Health Dashboard RSS feed: "Increased Connectivity issues in US-EAST-1", November 15, 2026 (read November 15)
- Computerworld: Amazon finds out AI programming isn't all it's cracked up to be (March 16, 2026)
The 30-second version
The FAA and Microsoft each said AI was not behind their outages. Both described a change that spread over one shared path. The Transportation Secretary says a router flooded the FAA's network during a copper-to-fiber line switch; officials now believe a feature that broadcast data too widely was involved, but the review has not announced a root cause. Microsoft's preliminary report traces its outage to a fix deployed from July that passed testing, sat latent, and could not be quickly rolled back. Add change control for ISP, firewall, and WAN work, including work your vendors do. Map your single points of failure. Write a continuity plan for when the cloud is up and email is not. Treat AI-assisted changes like any other change. If you'd like a senior engineer to walk through this with you, the project intake form takes about three minutes.
Pro IT NW helps regulated mid-market organizations run Microsoft 365, identity, security, and backup and disaster recovery. Vendor-neutral, labor-only. This post reflects public reporting and the FAA's own questions-and-answers page as of the afternoon of November 16, 2026 (Pacific), plus the AWS update added November 15. The FAA had not announced a root cause; Duffy called the Arkansas router the "likely cause," and L3Harris said it is supporting the investigation. Microsoft's preliminary post-incident report, dated November 15, was read from our own tenant's admin center; its final report is due within five business days. As of November 16, AWS had not published a post-event summary or a cause.
Questions we get asked
- What caused the FAA ground stop on November 13, 2026?
- It is not settled, though it has narrowed. Transportation Secretary Sean Duffy said the outage began while L3Harris Technologies, which runs the network, was switching lines from copper wire to fiber at a remote FAA site in El Dorado, Arkansas, and that a router there, and a second router in Amarillo, Texas, flooded the FAA's network with data. Both routers are being tested. In its own questions-and-answers page, dated November 15, the FAA says data flooded its FTI network "at a rate of 1,000 connection attempts every 4 to 5 seconds," names a radar site in Amarillo, Texas and a radio site in El Dorado, Arkansas as the two potential sources, and says it disabled the broadcasting function on the new routers. On November 16 Duffy called the Arkansas router the "likely cause" while FAA staff and FBI agents examined it at L3Harris's lab in Florida. The FAA says it and L3Harris "continue to investigate the root cause." L3Harris says it is supporting the FAA's investigation.
- Did AI cause the FAA or Microsoft outage?
- Both organizations said no. Bedford said there was "zero evidence of anything nefarious and zero evidence of anything AI related." Microsoft's public status page said the Microsoft 365 incident "was a technical issue not related to cyber or AI / agents." Both accounts are preliminary. The FAA review is open. Microsoft published a preliminary post-incident report on November 15, and its final report is due within five business days.
- What did Microsoft say caused the November 2026 Microsoft 365 outage?
- Microsoft's preliminary post-incident report traces incident MO1550821 to a latent defect in a fix prepared for Microsoft 365 network gateway servers, deployed from July 2026. It caused no impact for more than three months. A surge in load from a routine Teams failover, while nearly 14% of gateway capacity was out of service, triggered it, and the failure cascaded until more than 80% of gateway servers were unresponsive. Because the defect was months old, normal rollback could not quickly revert it, so Microsoft shipped an update that reverted the defective change. The final report is due within five business days.
- What is a single point of failure in IT?
- A single point of failure is one component, provider, or path whose loss stops a whole service, with no tested alternative behind it. The FAA's Bedford said all facilities rely on one network and "there's no backup to the backup." In a mid-market firm the usual candidates are one internet circuit, one firewall, one identity provider, one mail path, and one remote-management tool. A second copy only counts if it does not share the first one's failure, and if you have actually tested it.
- What should IT change management cover for network work?
- For any work on an internet circuit, firewall, or WAN link, write down a change window, what "done" looks like, a rollback plan, and who is physically on site. Keep the record somewhere searchable, because at the FAA, tracing the problem took about 45 minutes, which the FAA Administrator said was partly because maintenance is logged on paper. Apply the same rules to work performed by a vendor or contractor, not just your own staff.
- Does a business continuity plan cover a cloud outage when your own network is fine?
- It should, but many do not. Plan for the case where the cloud is up and one service is not, such as Exchange Online being unreachable while your office internet works fine. Write down how staff reach each other, which customer and vendor contacts live outside email, and who decides when to switch to the fallback. Microsoft's own status text during its November incident noted that service access can recover before mail flow does, so define what "back" means before you need it.
Related service
Backup & disaster recovery implementationWritten by the team at Pro IT NW · Senior-led Microsoft project consultancy · Seattle and the Pacific Northwest, delivered USA-wide.