InfrastructureCompute, chips & cloud
US officials say NSA, FAA, Microsoft, Chase and Apple outages were internal failures, not a cyberattack
Several major US outages in November were blamed on bugs, upgrades and, at the NSA, patches for PHP flaws that AI tools found. Officials ruled out a coordinated attack. A new guide shows how to tell outages from cyberattacks.
From the Ars Technica · AI article of November 19, 2026 and cited sources: Each outage has its own stated cause. [2] [3] [6] [8]
SOURCES[2] NSA outage disrupted communications — Jerusalem Post jpost.com[3] A week of outages — MLQ.ai mlq.ai[6] The Great Exchange Online Outage of November 2026 —… office365itpros.com
AI-assisted, not human-edited. Methodology Corrections Who owns this site
- Original source
- Ars Technica · AI article, independent perspective
- Brief by
- AT Pulse research desk AI-assisted; publisher and contact on the desk page
- When
- November 19, 2026 article published
- Covered
- Nov 19 edition
- Evidence
- 13 sources across 11 sites
- 1How to tell a cyberattack from a plain old outagearstechnica.com
- 2NSA outage disrupted communications — Jerusalem Postjpost.com
- 3A week of outages — MLQ.aimlq.ai
- 4PHP flaws found by AI — TechCrunchtechcrunch.com
- 5Websites shut down over PHP flaw — ABC7abc7news.com
- 6The Great Exchange Online Outage of November 2026 —…office365itpros.com
Reach: not measured yet

Each outage has its own stated cause. Microsoft blamed a gateway bug. The FAA blamed an overloaded router installed during a network upgrade. Chase cited a hardware software bug, and Apple an internal update. The Director of National Intelligence said the incidents were separate and not state-sponsored. The affected organizations or officials reported these causes themselves, and no independent forensic findings have been published. [2] [3] [6] [8] [9]
AI enters the story through finding vulnerabilities. Wiz researchers used AI agents to find long-standing flaws in PHP's core interpreter, which set off a rush to patch. Officials connected the NSA's two-day classified network outage to those PHP updates. Some experts tie the week's failures to rushed patching. The Transportation Secretary rejected any AI role in the FAA outage. [2] [4] [5] [8] [9]
In both the Microsoft and FAA outages, a small problem spread widely. For Microsoft, three things combined: a July Windows change that caused lock contention on servers with many CPU cores, Teams traffic failing over onto a gateway it shares with Exchange, and capacity that was already partly offline. Together they broke through the isolation between regions. For the FAA, one router failure during a copper-to-fiber migration spread across multiple air traffic control centers. [6] [8]
The triage guidance says to assume an accident first; one analyst estimates about one outage in a thousand is caused by hackers. Signals that help: routes or DNS disappearing, the geographic scope of the failure, and recovery time (a rollback takes hours, ransomware takes days or weeks). Public tools such as IODA, Cloudflare Radar and vendor status pages can confirm what is happening. [1] [10]
rushed changes and upgrades caused major outages, and officials found no sign of an attack. Implication: when AI speeds up the discovery of vulnerabilities, emergency patching itself becomes a risk to availability.
Read the full assessment
That makes staged rollouts, rollback paths and an outage-versus-attack playbook more important.
The research is done; here is the dossier.
Executive brief
A run of US outages over four business days had a defensive cause, not an attack. Officials tied the NSA's two-day classified network failure to software updates for PHP vulnerabilities that AI tools had found (Jerusalem Post). Microsoft, the FAA, JPMorgan Chase and Apple all blamed their own bugs, updates or upgrades. The Director of National Intelligence, Jay Clayton, said the outages were separate and not a coordinated or state-sponsored attack (MLQ.ai roundup). The WIRED/Ars Technica guide says to assume an accident first. Infoblox's Doug Madory puts hacker-caused outages at about one in a thousand.
What changed and event timeline
AI-found PHP flaws trigger a patching scramble
Wiz researchers used AI agents to find long-standing flaws in PHP's core interpreter. Some government websites went offline while operators patched (;).
Exchange Online fails worldwide
A bug in Microsoft's gateway, present since July, took down email, and the effects spread across Microsoft 365. Reported recovery times differ: about 24 hours in one account, Nov 14 in another (;).
FAA nationwide ground stop
An overloaded router installed in southern Arkansas during an L3Harris network upgrade overwhelmed radar and air traffic control data feeds. Departures stopped for more than four hours, and officials said there was no sign of a cyberattack ().
Capabilities and access
This story is not about a model release. The AI angle comes from vulnerability discovery. The Wiz team reportedly used its Wiz Atlas agent, Gemini 4 Argon and GPT-5.6 Sol to find the PHP flaws (TechCrunch).
Read the full section
This story is not about a model release. The AI angle comes from vulnerability discovery. The Wiz team reportedly used its Wiz Atlas agent, Gemini 4 Argon and GPT-5.6 Sol to find the PHP flaws (TechCrunch). The guide relies on public tools for diagnosis:
- Georgia Tech's IODA outage detection service
- Cloudflare Radar
- Watch the Net
- Vendor status dashboards for AWS, Google Cloud and Azure (Ars Technica)
Technical analysis for researchers and developers
The isolation boundaries between regions did not stop the failure from spreading, and backlogged traffic slowed recovery.
Read the full section
- Microsoft: Three conditions combined in its preliminary review (Office365itpros):
- a Windows change from July caused lock contention on newer servers with many CPU cores, which testing missed;
- Teams had failed over onto a gateway it shares with Exchange;
- about 14% of gateway capacity was already offline.
The isolation boundaries between regions did not stop the failure from spreading, and backlogged traffic slowed recovery.
- FAA: One router failure spread across multiple control centers while the network was being moved from copper to fiber (CNBC).
- Triage signals:
- Routes or DNS that disappear usually point to the company's own change.
- Scope matters: one provider failing across countries suggests a bad update, one region suggests infrastructure, one country suggests government action.
- Recovery time matters: a rollback takes minutes to hours, ransomware takes days to weeks (IPinfo).
Claims and evidence
- Not an attack: Clayton said he had "a very high level of confidence" the outages were not cyberattacks, and Secretary of State Marco Rubio said the US was not under systemic attack (Jerusalem Post).
- Causes are mostly self-reported. No independent forensic reports have been published (MLQ.ai).
- Durations conflict between sources:
Read the full section
- Not an attack: Clayton said he had "a very high level of confidence" the outages were not cyberattacks, and Secretary of State Marco Rubio said the US was not under systemic attack (Jerusalem Post).
- Causes are mostly self-reported. Each cause comes from the affected organization or officials: Microsoft's preliminary report, Transportation Secretary Sean Duffy, and statements from Chase and Apple. No independent forensic reports have been published (MLQ.ai).
- Durations conflict between sources:
- Chase: about 12 hours (NBC Palm Springs) or about 9.5 hours (MLQ.ai).
- United: 90 minutes or about 4 hours, in the same two sources.
- The "one in a thousand" figure is Madory's own estimate (Ars Technica).
Context and prior work
- The FAA ground stop was the longest halt to US departures since 9/11 (CNBC). The FAA had a previous system outage in 2023 (Wikipedia).
- Past attacks looked different. The 2016 Mirai attack on Dyn came with no claim of responsibility (Ars Technica).
- Real attacks did happen in the same period.
Read the full section
- The FAA ground stop was the longest halt to US departures since 9/11 (CNBC). The FAA had a previous system outage in 2023 (Wikipedia).
- Past attacks looked different. NotPetya and WannaCry left ransom notes on infected machines. The 2016 Mirai attack on Dyn came with no claim of responsibility (Ars Technica).
- Real attacks did happen in the same period. The Optimizely supply-chain compromise hit 178 sites, and a cyberattack forced refrigeration at French supermarkets into defrost (IPinfo).
- In May, lawmakers pressed the White House to respond to AI models that find vulnerabilities faster than they can be patched (Axios).
Limitations, safety and contested findings
- AI's role is disputed. Duffy called suggestions that AI tools caused the FAA outage "completely unfounded" (CNBC).
- The FAA's root cause is still under investigation.
- The NSA link to the PHP patches comes from officials speaking anonymously or to Congress.
Read the full section
- AI's role is disputed. Duffy called suggestions that AI tools caused the FAA outage "completely unfounded" (CNBC). CNBC's own reporting, by contrast, cites experts who link the week's failures to patching under pressure from AI-accelerated vulnerability discovery (CNBC).
- The FAA's root cause is still under investigation.
- The NSA link to the PHP patches comes from officials speaking anonymously or to Congress. It has not been published as a technical finding (Jerusalem Post).
- Hacker claims are unreliable. Attacker groups exaggerate or lie, and confirmation from investigators can take weeks to months (Ars Technica).
Business and practitioner implications
- Patching speed is now a risk to availability. Use staged rollouts and keep rollback paths ready.
- Microsoft's failure spread because the boundaries between gateway regions did not hold, and shared dependencies like the Teams/Exchange gateway amplify problems.
- Check routes and DNS, scope, timing against maintenance windows, and vendor status pages before declaring an incident (IPinfo).
Read the full section
- Patching speed is now a risk to availability. AI-found vulnerabilities push teams into emergency changes, and those changes caused outages in their own right. Use staged rollouts and keep rollback paths ready.
- Test isolation boundaries. Microsoft's failure spread because the boundaries between gateway regions did not hold, and shared dependencies like the Teams/Exchange gateway amplify problems.
- Prepare an outage-versus-attack triage playbook. Check routes and DNS, scope, timing against maintenance windows, and vendor status pages before declaring an incident (IPinfo).
- Communicate early. Saying "investigating" for a long time feeds speculation that it is an attack.
The source trail.
Sources (13)
How to tell a cyberattack from a plain old outage
Article text retrieved; extracted text may omit tables or interactive elements.
arstechnica.com