The Tech Meltdown Nobody Hacked: What Last Week’s Outages Really Tell Us

The Tech Meltdown Nobody Hacked: logos of Microsoft, FAA, JPMorgan Chase, United, Apple, Marriott and NSA with warning signs. Major outages. No confirmed attack.

By Kate Eye, CPA, CISA, Founder, Eye & Co.

Last week was a tech meltdown, and plenty of people online were ready to call it IT doomsday. Microsoft email went down, the FAA stopped every departure in the country, then a big bank, an airline, Apple, Marriott, and an NSA network all had problems within days of each other. The theories came fast: a coordinated attack, rogue AI, the beginning of the end of days…

So far, the evidence points somewhere much less dramatic. Every affected company or agency has said its problem was not the work of an outside actor (BBC). Secretary of State Marco Rubio put it plainly: "We are not under systemic attack" (BBC). The explanations we have so far are a months-old software bug, a router installed on a contractor's network, internal updates, and routine maintenance.

Which raises a fair question: aren't there supposed to be controls in place to prevent this kind of thing? These are large, sophisticated organizations, and they likely have external auditors along with their own risk, compliance, and security teams. The companies on that list almost certainly have SOC 2 or similar reports covering their services, and those reports are likely clean. (The FAA and NSA are government agencies, so they work under different frameworks.) I don't know what any of those reports say. But I do know this: a clean report isn't the same as no risk.

A SOC 2 can be a badge you post on your trust page, or it can be a real second look at how your systems fail. The difference is audit quality. A rubber-stamp audit asks whether a policy exists. A quality audit asks how you actually handle a security incident, whether failover has actually been tested, and which vendor change could take you down. Last week is a good example of why those questions matter.

What happened, Nov. 12–17

Here's the week, using only what companies, agencies, and reporters have confirmed.

  • Thursday, Nov. 12: Microsoft 365. Outlook, Exchange Online, Teams, and Copilot were disrupted worldwide from 11:20 a.m. until 11:10 a.m. the next day (BleepingComputer). Microsoft's preliminary report names three factors working together: a latent defect in a gateway fix that began rolling out in July and wasn't caught by load testing, extra load after Teams failed over to the gateway that also serves Exchange Online about 15 hours earlier, and nearly 14% of gateway capacity already out of service from an unrelated issue. As servers froze, traffic was rerouted to healthy ones, which then froze too. At the low point, more than 80% were unresponsive (BleepingComputer; Ars Technica).

  • Friday, Nov. 13: FAA nationwide ground stop. Data flooded the FAA's telecommunications network at about 1,000 connection attempts every four to five seconds, destabilizing legacy equipment and causing instability at 17 of the FAA's 22 en route centers. Investigators flagged two sites: a radar site in Amarillo, here in Texas, and a radio site in El Dorado, Arkansas, where L3Harris, the FAA's network service provider, had been deploying a new router. Flights resumed at 8:15 p.m., and operators canceled about 12,462 flights that day (FAA). It was the longest halt to U.S. departures since 9/11 (CNBC).

  • Monday, Nov. 16: JPMorgan Chase and Zelle. Chase blamed "a software bug in on-premise hardware systems" that slowed transactions, and said other banks were affected only when they were party to a transaction with Chase (CNBC). Customers of Bank of America, Wells Fargo, and U.S. Bank also reported problems, and some Zelle payments were delayed (BBC).

  • Monday, Nov. 16: United Airlines. Pilot-facing applications went down, so crews couldn't get the final aircraft weights they need to calculate takeoff performance. The fix came about an hour and 40 minutes after the first pilot reports. United said there was no indication of a cybersecurity incident but didn't say what caused it (AeroTime).

  • Tuesday, Nov. 17: Apple. iCloud, Apple Pay, and Apple's retail store systems were disrupted. Apple blamed an unspecified "internal update" and said it was not a cyberattack (MacRumors; CBS News).

  • Tuesday, Nov. 17: Marriott. Sign-in failed on Marriott.com and the Bonvoy app for several hours, which also cut off mobile check-in and digital room keys. Marriott said its teams worked with its technology partners on the fix and found no indication of a cybersecurity cause (Skift).

  • Tuesday, Nov. 17: NSA. Some of the agency's internal systems went down during what it called "scheduled routine maintenance" (Politico).

Two caveats as I write this on Nov. 18. The FAA's root cause is still officially open. As of Nov. 17, Transportation Secretary Sean Duffy said, "The root cause of what happened, we don't have clarity yet" (Los Angeles Times). And Microsoft's final post-incident report wasn't out yet, so the details above come from its preliminary report (BleepingComputer). I won't guess past what's been confirmed.

If you're wondering why so many people blamed AI first, and what your customers are starting to ask about your own AI, I covered that in my last article on AI and SOC 2.

What the week actually has in common

No one has reported a shared cause. "Humans see patterns in everything, regardless of whether they're there, and generally interpret coincidences as conspiracies," security technologist Bruce Schneier told Axios (Axios). But there is a shared pattern. In almost every case, a change (an update, an installation, maintenance, a failover) met conditions nobody had tested for, and the problem spread through something other systems depended on.

That's not a new kind of failure. It's the oldest kind. David Woods, a professor emeritus at Ohio State, described what happens when complex systems are pushed to move faster and cheaper: "We're cutting out the cross checks, we're cutting out the tests, we're cutting out the stress tests, we're cutting out the safeguards" (Axios). An audit is one of the few cross checks most growing companies get every year. It's worth making it count.

Rubber stamp vs. quality audit

I want to be careful here. I'm not saying a better audit would have prevented any of these outages. Large, complex systems fail, and no audit sees everything. What a quality audit gives you is a chance to catch a weak spot before your customers do.

Here's the difference in practice:

  • A rubber-stamp audit checks that things exist. Is there a change management policy? Is there a vendor list? Is there a disaster recovery plan? Did someone sign off?

  • A quality audit checks whether those things work. Do your test conditions match production? Has failover been tested under realistic load? Which vendor change could take you down, and would you know before your customers did? Do the reports you rely on cover the system you actually depend on?

Both can end in a clean report. Only one leaves you understanding your risks better than when you started. Change management, vendors, and availability are where last week's outages show the gap most clearly.

What a quality audit probes: change management

In SOC 2, the change management criterion, CC8.1, asks whether you authorize, design, configure, document, test, approve, and implement changes to infrastructure, data, software, and procedures (AICPA). Notice that it says infrastructure, not just code. A router swap, a firewall rule, a DNS update, a gateway setting: those are all changes. In my experience, network and infrastructure changes (unless in code) are the ones most likely to skip the process, because they feel like maintenance.

A rubber stamp confirms there's a ticket and an approval. A quality audit also asks the questions last week raised:

  • Does testing look like production? Microsoft said its load testing missed the defect because its labs focused on weaker hardware with fewer cores than some of its production machines (BleepingComputer). Senate Commerce Committee Democrats asked whether the FAA's testing "adequately reflected operating conditions" (U.S. Senate Commerce Committee).

  • Is the rollout staged and watched? By Duffy's account, the FAA's problem came on the 69th router installation: "68 had gone really well" (ABC News). Microsoft's change ran for more than three months before the right conditions exposed it (BleepingComputer). Staging helps, but it doesn't replace monitoring and a way back.

  • Is there a real rollback plan? The senators asked the FAA to demonstrate "verified failover capability, and rollback procedures" before deployment resumes (U.S. Senate Commerce Committee).

  • Is the change window chosen on purpose? The FAA said maintenance at major facilities is typically done at night, and that it will "reevaluate" when it does this kind of work (FAA).

  • Is configuration treated as part of the change? HPE, which owns the router maker, Juniper, said, "Based on what we know and have been told by others, our equipment was working as configured" (ABC News). Whatever the investigation finds, your process should cover the settings, not just the box.

There's also a reason this may keep happening for a while. In October, a cluster of serious security flaws was disclosed, including two in PHP that allow remote code execution on web servers and that researchers found with the help of AI. The security world now calls that stretch "Black October" (CodexGuild; INCIBE-CERT). Sen. Mark Warner said AI is creating "enormous pressure on companies and government agencies to patch those vulnerabilities immediately, often all at once" (Los Angeles Times), and a senior administration official attributed last week's incidents to "upgrades of the systems, systems being patched, hardware being applied" (CNBC). You need to patch fast, and every patch is a change. A good auditor will ask how your emergency changes work: who can approve an urgent patch, what minimum testing still happens, and how you roll back.

What a quality audit probes: vendors

The FAA didn't install that router. Its contractor did, on a telecommunications network the contractor owns and operates (KATV). The FAA still owns the consequences: the canceled flights, the press conferences, and a letter from senators asking for answers. Marriott needed its technology partners to restore sign-in (Skift). Chase's bug reached customers of other banks through shared transactions (CNBC).

CC9.2 asks whether you assess and manage the risks that come from vendors and business partners (AICPA). A rubber stamp confirms you have a vendor list. A quality audit asks which vendors could take you down, what you've required of them, and how you check. This isn't just a big-company problem. Over the nine years Uptime Institute has tracked publicly reported outages, third-party IT and data center service providers, including cloud, internet, telecommunications, and colocation companies, have accounted for about two-thirds of them (Uptime Institute).

SOC 2 has specific language for this, and it's worth knowing:

  • Carve-outs. When your system relies on a vendor such as a cloud host or managed network provider, your SOC 2 report often "carves out" that vendor as a subservice organization. Their controls aren't tested in your report, but your report has to describe what you're relying on them to do (AICPA DC 200). Your auditor should also look at how you monitor them.

  • Complementary user entity controls (CUECs). These are the flip side: the things a vendor's SOC 2 report assumes you, the customer, are doing (AICPA DC 200). If a vendor's report assumes customers review and approve changes to their own configurations, and you don't, part of the assurance you're counting on isn't there.

Then there's concentration. A handful of providers, including Microsoft, Amazon, and Google, run a huge share of the world's servers (BBC). Regulators have noticed. One year ago, on Nov. 18, 2025, EU financial regulators published their first list of critical tech providers under the Digital Operational Resilience Act (DORA), putting those providers under direct oversight (ESMA). Cloudflare's major outage, triggered by a database permissions change, happened that same day (Cloudflare). For a growing company, this shows up in customer security questionnaires: Which providers do you depend on? What happens if one goes down? When did you last test recovery? After a week like this one, I'd expect more of those questions, and more specific ones.

What a quality audit probes: availability

If your customers care about uptime, the Availability criteria are where capacity monitoring, backup and recovery infrastructure, and recovery testing live (AICPA). Microsoft's outage is a good reminder that failover has to be tested under realistic load: a routine Teams failover added the traffic that helped expose the bug, and rerouting to healthy servers spread the failure instead of containing it (BleepingComputer). A rubber stamp confirms there's a recovery plan. A quality audit asks when it was last exercised, under what conditions, and what you learned.

There's a catch, though. In my experience, many SaaS SOC 2 reports cover the Security criteria only. If Availability isn't in scope, none of this gets tested. That's not automatically wrong, but it should be a deliberate choice that matches what your customers depend on.

A badge isn't a scope

That brings me to the badge itself. When something like this happens, the first instinct is to check whether a vendor "has a SOC 2." That's the right question, but it's only the start. Reports are real, but they often cover a specific product, entity, or service, which may not be the system you depend on.

That's why I tell clients to read the scope before the conclusion. When you get a vendor's SOC 2 report, check which system and services are in scope, which criteria were covered (Security only, or Availability too), the period, which subservice organizations are carved out, which CUECs land on you, and any exceptions in the testing. A logo on a trust page tells you a report exists. It doesn't tell you the report covers the thing you depend on. The same goes for your own report.

What your auditor should actually test

Here's where this gets concrete. In a quality SOC 2 audit, these are the areas where last week's lessons show up as evidence:

  • Change management (CC8.1). Your auditor pulls a sample of real changes from the audit period, including infrastructure and configuration changes, not just code. For each one, they look for approval, evidence of testing, and who put it into production.

  • Vendor and subservice oversight (CC9.2). Your auditor looks at how you identify and rate vendors, whether you actually reviewed your critical vendors' SOC 2 reports, how carved-out subservice organizations are described in your system description, and whether you perform the CUECs those reports assign to you.

  • Availability (A1.2 and A1.3). If Availability is in scope, your auditor looks for evidence of backups, recovery infrastructure, and recovery testing: restores that were actually performed, failover that was actually exercised, and documented results.

  • Incident response (CC7.3 to CC7.5). Your auditor looks for evidence that security events were evaluated, incidents were handled according to your plan, and recovery was carried out: tickets, timelines, communications, and post-incident reviews with follow-up actions.

A good audit tests real changes and real recoveries, not just whether a policy exists. A policy that says "all changes are tested" isn't evidence; the change tickets and logs in the source code repository are.

What you should get beyond the report

A quality audit should leave you with more than an opinion letter. Here's what that can look like:

  • Actionable findings. If a sample turns up exceptions, you should understand the root cause and what to fix, not just see a line in the report. An auditor's exception is a much cheaper early warning than an outage.

  • Observations beyond formal exceptions. Some of the most useful feedback doesn't rise to a formal exception: a process that technically passed but is fragile, a vendor that's become a single point of failure, a test environment drifting from production. A good auditor shares these.

  • A partner between audits. Risks don't wait for your next audit period. When you're planning a big infrastructure change, adding a critical vendor, or answering a tough customer questionnaire, you should be able to ask your auditor how it affects your controls.

None of this guarantees you won't have an outage. It does mean more people are looking for the weak spot, and more chances to find it before your customers do.

5 things to check before your next big change

  • Put every infrastructure change through the process. Network, DNS, firewall, and cloud configuration changes get an approver and evidence of testing, the same as code. Leveraging an infrastructure as code approach is a great way to enforce this.

  • Roll out big changes in stages, with a written rollback. Start small, watch the metrics, and decide in advance who can call "stop" and how long it takes to undo. Have an emergency change path for urgent patches, too.

  • Tier your vendors by what they can take down. For the ones that could cause an outage, know what changes they make in your environment, how they notify you, and when they're allowed to make them. That's CC9.2 in practice.

  • Read vendor SOC 2 reports instead of filing them. Confirm the in-scope system and criteria, note the carve-outs, and give every CUEC an owner on your team.

  • Test availability instead of assuming it. Exercise capacity alerts, failover, and restores on a schedule, under realistic load where you can, and keep the results. That's what the Availability criteria are asking for.

The quiet advantage

Within days, Microsoft published a preliminary explanation that walked through three specific factors (ABC News). Five days after the ground stop, the FAA's root cause is still open. The difference between those two situations lives in change records, test results, configurations, and vendor documentation. Most growing software, healthtech, fintech, and MSP companies can have those records ready before anything breaks, and a good audit is a yearly chance to pressure-test them.

Your SOC 2 should leave you smarter about your risks, not just holding a report. If yours didn't, let's talk. I'm happy to be a second set of smart eyes. Book 15 minutes with me at cal.com/kate-eye.

Kate Eye, CPA, CISA, is the founder of Eye & Co., a boutique SOC 2 audit firm in Austin, Texas, working with growing B2B software, healthtech, fintech, and MSP companies.

Sources

Next
Next

Customers are asking how you’re using AI. Can your SOC 2 answer?