Why US tech outages keep happening: a week of failures, explained
In six days a Microsoft bug, an FAA router, bank systems, Apple and an NSA network all failed. Officials say there was no cyberattack. The real story of these US tech outages may be ageing infrastructure meeting a flood of AI-found software flaws.

At 3:54 p.m. Eastern time on Friday, November 13th, the Federal Aviation Administration did something it had done only a handful of times in its history: it stopped every departure in the country. Controllers at the Chicago en route centre had been watching radar returns flicker and radio links fill with static. The fault spread to Boston, Washington, New York and Minneapolis. When the ground stop was lifted four hours and 21 minutes later, it had become, by CNBC's reckoning, the longest halt to American departures since September 11th 2001. Bryan Bedford, the FAA administrator, said about 12,000 flights were never completed.
The culprit, according to Transportation Secretary Sean Duffy, was a single router. It had been installed at a remote site in southern Arkansas by L3Harris, the contractor that runs the FAA's telecoms backbone, as part of a switch from copper to fibre. Once switched on, it fired off more than 1,000 connection attempts every four to five seconds and swamped the equipment that carries radar and voice data between facilities.
On its own the episode would have been an aviation story. It was not on its own. The day before, Microsoft 365 had gone dark for much of the world. Three days later customers of JPMorgan Chase and several other big banks could not see their accounts, and United Airlines halted departures over a software failure. On Tuesday Apple Pay and iCloud stumbled, and the National Security Agency's classified internal network went down. By Wednesday "why does everything keep breaking" had become a question asked by senators, not just by people stuck at gates.
Why the wave of US tech outages matters now
The official answer, delivered with unusual insistence, is that nothing connects the incidents. "We have no indication, no evidence, and no reason to believe at all that any of these incidents are related to each other," Secretary of State Marco Rubio told reporters on Tuesday. "The FAA has nothing to do with what we saw with the banks a few days ago." Jay Clayton, the director of national intelligence, said on Fox Business: "We do not see the indicia of a coordinated cyber attack." Every company involved has said the same about its own systems.
Yet the second half of Mr Clayton's remarks pointed somewhere more interesting. "Lots of institutions, because these vulnerabilities have been identified, are patching. They're taking their systems down," he said. Of the NSA, he was specific: the patching that followed a recently disclosed flaw in PHP, a programming language that runs a large share of the web, "has caused some of these systems to not come back up when you patch them as expected. That is what has happened at the NSA today." The agency itself described the incident only as "temporary degradation during scheduled routine maintenance".
That framing matters because it shifts the debate from a hunt for a hidden enemy to a question about how fast critical systems can be changed safely. The stakes are not abstract. Bernstein, a research firm, estimates that the FAA stop alone cost American, Delta, United and Southwest about $345 million in pre-tax earnings; Raymond James puts the hit to US carriers above $1 billion. Thanksgiving, the busiest travel week of the American year, begins on November 26th.
Six days, six failures
Strip away the speculation and the record of each incident, as stated by the organisations themselves, looks like this.
Microsoft's outage began on November 12th and hit Outlook, Teams and other services in more than 70 countries. In a preliminary review the company blamed a latent defect in the gateway servers that front Microsoft 365, dormant since an update in July 2026, which surfaced when maintenance shifted traffic and capacity was reduced. "This was a technical issue not related to cyber or AI/agents," Microsoft said. A final report is due by November 20th.
The FAA failure came next. The router was made by Juniper Networks, now part of Hewlett Packard Enterprise, which said the device "functioned as it was configured to by the installer". Mr Duffy has not hidden where he thinks the problem lies: "I'm stuck with L3Harris, who owns this network." Investigators are still trying to determine whether the cause was a faulty unit, a misconfiguration or an installation error. On November 17th the FAA paused installation of new routers and disabled a function on dozens of identical devices already in the field.
JPMorgan Chase said its November 16th disruption was "a temporary issue related to a software bug in on-premise hardware systems" and that "there is no evidence this is cyber-related". Bank of America, U.S. Bank and Capital One customers also reported problems, though several of those banks said their own systems were not directly affected. Apple attributed Tuesday's failures to an internal update. President Donald Trump, asked about the NSA in Manila, said: "Things like that happen, it's fine."
There was one genuine attack in the same week, which has fed the anxiety. Attackers exploited a flaw at Optimizely, a web optimisation vendor, to plant malicious code on well-known websites, among them HSBC, the BBC and the crypto-wallet maker Ledger. Optimizely said "an unauthorized party exploited a vulnerability" and that it had removed the injected code. Nobody in government has linked that breach to the outages.
The AI patch rush and "Black October"
To understand Mr Clayton's remarks, rewind to October 21st. That day PHP's maintainers shipped emergency releases fixing two critical flaws, CVE-2026-109618 and CVE-2026-109619, collectively dubbed PHP2Shell and rated 9.2 out of 10 for severity. According to Wiz, the cloud-security firm that reported them, nearly every version of PHP released in the past decade was vulnerable, and even a bare "Hello World" application could be taken over remotely. Palo Alto Networks' Unit 42 counted 22.6 million potentially vulnerable instances. PHP told site owners to patch immediately or take their websites offline.
What made the episode different was how the bugs were found. Researchers at Calif, a security start-up, spotted the first issue on August 1st using OpenAI's GPT-5.6 Sol model. Wiz found it independently weeks later with help from OpenAI's and Google's frontier models. "With frontier AI models and the right tools, a single researcher completed in a week what previously would have taken a world-class team months of work," Wiz's researchers wrote. OpenAI, which coordinated the disclosure, called it "one of the most impactful security issues in recent memory" and warned that "the same capabilities can also help attackers rapidly develop offensive tools and exploits".
Attackers did not wait. Matthew Prince, Cloudflare's chief executive, reported tens of thousands of exploit attempts within 45 minutes of the news breaking and millions by the evening of October 21st. CISA, the federal cyber-defence agency, added both flaws to its catalogue of actively exploited vulnerabilities on October 26th. Several governments, including agencies in the Netherlands and Finland, pulled websites offline as a precaution.
PHP2Shell was the loudest example of a trend that has been building all year. In April Anthropic said its Mythos model had found thousands of previously unknown vulnerabilities across major operating systems and browsers; Treasury Secretary Scott Bessent and Federal Reserve Chair Jerome Powell then summoned bank chief executives to discuss the risk. The logic of the past few months is simple. If machines can find flaws faster than ever, defenders must patch faster than ever, and every patch is a change to a live system.
Senator Mark Warner of Virginia, the vice-chairman of the Senate Intelligence Committee, made that the centre of his reading of the week. "What appears to be happening instead is a consequence of how quickly artificial intelligence is changing the cybersecurity landscape," he said in a statement. "AI is making it possible to identify vulnerabilities across enormous numbers of systems at unprecedented speed," creating pressure "to patch those vulnerabilities immediately, often all at once."
Old pipes, new pressure
The trouble with a single grand theory is that the incidents with the most public detail do not fit it neatly. The FAA's router was not a security patch; it was a hardware swap in a modernisation programme. Mr Bedford said there was "zero evidence of anything AI-related", and Mr Duffy called the idea that AI tools had a role "completely unfounded". Microsoft's bug had been sitting quietly since July.
What the FAA case does show is how old the plumbing is. The FAA Telecommunications Infrastructure network, which links more than 4,000 facilities, has been flagged as an outage risk by federal auditors for more than 15 years. "Every agency within the government for better than two decades has flagged this FTI system as a critical weakness," Mr Bedford said on November 14th. "Let's not kid ourselves; it will continue to break until we replace it." In September 2024 the Government Accountability Office found that 51 of the FAA's 138 air traffic systems were unsustainable and 54 more potentially so.
There is even a precedent. On November 19th 2009 a temporary router that an engineer for Harris Corporation, L3Harris's predecessor, had configured incorrectly blocked about three-quarters of the network's routes and disrupted 129 air traffic facilities. The Transportation Department's inspector general concluded at the time that the FAA had been "reactive, rather than proactive". Seventeen years later the agency is again replacing copper with fibre, and again a router has brought the system down.
The GAO also warned, in a report published on September 14th 2026, that the FAA had pulled forward its target for replacing 5,170 copper lines and moving nearly 30,000 services to internet protocol by 21 months, to December 2028. By May only 2,560 lines had been replaced. "In essence, the FAA is buying delivery speed at the risk of cost uncertainty and potential disruptions," the auditors wrote. Two months later that risk materialised.
Markwayne Mullin, the homeland security secretary, offered a version of the same diagnosis for the private-sector failures: newer technology being layered onto older computer systems that were not designed to handle it. Monika Akbar, a cybersecurity researcher at the University of Texas at El Paso, put it more modestly. "It could be pure coincidence. All systems go through regular maintenance and updates," she told KVIA, an ABC affiliate. But she suggested that the rush to automate code management and software updates "might need to slow down or have more people in the loop".
Where the experts split
Few serious voices claim to have evidence of sabotage. The disagreement is over how much benefit of the doubt to give. Ram Dantu, founding director of the Center for Cyber and AI Security at the University of North Texas and a former Cisco director, said doing network upgrades in the middle of a working day was "strange". He accepted that congestion or bugs could explain the failures, but warned that an adversary could plant malware that "will keep quiet for a long period of time". "I think we are showing our weakness and I am really concerned about it," he told Fox News.
Christiaan Beek, a senior director at the security firm Rapid7, took the opposite view in comments reported during the week: "Complex systems fail in boring ways far more often than because of an adversary." He nonetheless added that outdated change processes become riskier as AI-assisted research increases the volume of flaws to fix.
On Capitol Hill the split runs partly along party lines and partly across them. Senator Rand Paul, the Kentucky Republican who chairs the Homeland Security committee, said there was "no evidence that there was a foreign actor or a cyberattack, but we certainly should look". Jim Himes, the senior Democrat on the House Intelligence Committee, called it a "deep breath moment". His fellow Democrat Seth Moulton was blunter: "I don't trust this administration to tell us the truth." Democrats on the Senate Commerce Committee, demanding a briefing from the FAA, L3Harris and Juniper, noted that "identifying the device that failed does not establish why it failed".
The data cut both ways too. Downdetector figures cited by The Washington Post show that the seven days to November 16th produced the highest weekly count of outage reports in the past 12 months, but that the picture over 30 days is far from unprecedented. A week can look like a pattern and still be a cluster.
Money is the other fault line. Mr Duffy has asked Congress for $50 billion to rebuild air traffic control, on top of the $12.5 billion approved in 2025, and warned on CNN: "You will have more of these." Ted Cruz, the Republican chairman of the Senate Commerce Committee, is sceptical. "I don't think the full $50 billion will be funded. Some of that is not safety related," he said.
Scenarios: what could happen next
The benign scenario is the one officials are betting on. Each incident has a mundane, documented cause; the cluster fades; post-incident reports from Microsoft and the FAA confirm configuration errors and latent bugs; and the week becomes a case study in change management rather than in national security.
The middle scenario is the one Mr Warner and Mr Clayton have sketched. The volume of AI-discovered vulnerabilities keeps rising, regulators and security agencies keep urging urgent patches, and organisations running brittle legacy systems keep breaking them while fixing them. Outages become a recurring cost of a security arms race, concentrated around big disclosure dates. Mr Clayton himself said: "We will have other patches going forward as we upgrade our infrastructure."
The darkest scenario, which nobody in government endorses and no public evidence supports, is that an adversary exploits the noise. Even officials who dismiss a coordinated attack concede that the flood of exploit attempts after PHP2Shell was real. A system taken offline for patching is safe; a system left unpatched because its owners fear another outage is not. That trade-off, rather than any single router, is the uncomfortable choice the week has exposed.
What to watch
Several markers will show which scenario is unfolding. Microsoft's final post-incident review is due by November 20th and should say whether the July defect was caught by testing that failed or by none at all. The FAA's root-cause analysis, with L3Harris, Juniper and the FBI examining the Arkansas hardware, will determine whether router installations resume before the Thanksgiving rush that peaks around November 26th. The Senate Commerce and House Transportation committees have both promised reviews of the ground stop.
Beyond that, watch the patch calendar. Microsoft's next scheduled security update falls on December 8th, the first big Patch Tuesday since the outages, and any stumbles there will be read as evidence for Mr Warner's thesis. Watch, too, how quickly the 22.6 million exposed PHP instances shrink, and whether the $50 billion request survives the spending fights in Congress. The FAA's network replacement is due to be finished by December 2028. Until then, by the agency's own admission, it will keep breaking.
This article is for information purposes only and does not constitute investment advice.



