SDSignal Desk

A week of outages, no attack, and a patching problem AI made worse

· Updated · Signal Desk newsroom

Analysis of reporting by WIRED. How we work

Image: WIRED

Microsoft, the FAA, Chase, United, the NSA and Apple all stumbled in one week. Officials say none were cyberattacks. WIRED's Uncanny Valley makes the case that the likelier culprit is a rush of fixes, some of them found by AI.

Key takeaways

  • Within a week, Outlook and Teams went down for about a day, the FAA ordered a nationwide ground stop of more than four hours, and Chase, United, the NSA and Apple all reported failures.
  • Every organization involved has said its incident was not a cyberattack, and WIRED's hosts found no evidence the failures were linked, apart from banks as counterparties.
  • The stated causes are mundane: a latent Microsoft bug from a July update, aging FAA comms equipment during a Friday update, bugs, maintenance and internal updates.
  • The hosts argue AI still plays a role, because vulnerability-hunting models are pushing out patches faster than systems can safely absorb them.
  • Public suspicion is fed by low trust in official reassurances, not just by fear of AI.

Everything broke at once

The list reads like a stress test. In WIRED's telling on its Uncanny Valley podcast, Microsoft's Outlook and Teams were down for roughly 24 hours. The next day the FAA ordered a nationwide ground stop for more than four hours, which the hosts describe as only the second since 9/11 and the longest since then. Chase then had a problem that rippled across banks, United delayed more than 100 flights for an unrelated reason, the NSA's classified internal network went down, and an internal Apple update knocked out iCloud and Apple Pay for a shorter stretch.

Each organization has said its incident was not a cyberattack, and senior business editor Sophie Kleeman, Brian Barrett and Leah Feiger say there is no evidence these failures were connected beyond the banking chain. That has not stopped people from assuming hackers or rogue AI. The episode's useful contribution is to take that instinct seriously instead of mocking it.

The boring explanations

The causes, as described, are humdrum. Microsoft traced its outage to a change deployed in July that sat dormant until a later update triggered it. Satya Nadella separately called it a latent bug that led to a perfect storm, which we covered in his comments on unpatchable infrastructure. The FAA failure was a communications problem tied to old fiber equipment maintained by L3Harris, which was being updated on a Friday afternoon. Chase blamed a software bug, United a technology issue, the NSA routine maintenance and Apple an internal update.

The FAA case shows how coincidence fuels suspicion. The agency had rolled out an AI system for air traffic data the day before, so the outage looked connected. According to WIRED, the failing system handled voice communications, not the new data tool. Timing also matters: year-end pushes before holiday travel and banking seasons concentrate risky changes in the same weeks.

Where AI does fit

The most interesting argument in the episode is that the conspiracy theorists are wrong about the mechanism but not entirely wrong about AI. Barrett points to Anthropic's Mythos model, which he describes as very good at finding security flaws and which has been made available to a sizable set of governments and companies. Those organizations are finding bugs they could not find before and pushing patches at a high rate, and the more complex the fix, the more likely something breaks. We treat that as a plausible hypothesis the hosts attribute to reporting, not a confirmed root cause for any single outage.

It is a real shift. For decades, defenders had a backlog of unknown bugs and time to test fixes. Now discovery is cheap and fast, and the bottleneck moves to safe deployment. Nadella made the same point: when a flaw is being exploited in the wild, a patch must go straight to production, and that creates instability. The industry has traded one risk, unpatched systems, for another, rushed change. Neither is free, and the right answer is better staging and rollback, not slower discovery.

Trust is the other outage

The hosts are frank that skepticism is also about who is doing the reassuring. Public statements from officials, including Marco Rubio's insistence that the country is not under systemic attack, landed with an audience that does not trust them. Kleeman adds that AI companies have done a poor job of making people comfortable with their products, so AI becomes the default suspect. When institutions want calm, they need fast, detailed postmortems. Microsoft has promised a longer one. That document, and the FAA's, will do more to settle nerves than any press conference.

The episode also covers Anthropic's IPO, Nvidia's first earnings since Jensen Huang's resignation and a new rule on Chinese companies renting AI chips; for the IPO side, see our coverage of Anthropic's reputation test.

Who feels it

IT and ops teams
Faster vulnerability discovery means more urgent patches and more pressure on change management.
Travelers and bank customers
They absorb the cost of risky year-end changes in missed flights and failed payments.
AI security vendors
Their tools are blamed for chaos even when they are helping defenders.
Regulators
Clustered failures in critical services invite questions about deployment practices, not just security.

What to watch

  1. Microsoft's full postmortem on the Outlook and Teams outage.
  2. Whether the FAA or L3Harris changes its update windows for legacy comms equipment.
  3. Any evidence linking specific outages to AI-discovered vulnerability patches.
  4. Industry guidance on staged rollouts for emergency patches.

FAQ

Were last week's outages caused by cyberattacks?

Microsoft, the FAA, Chase, United, the NSA and Apple all said their incidents were not cyberattacks, and WIRED's hosts found no evidence they were linked beyond the banking chain.

What caused the FAA ground stop?

According to WIRED, a communications failure tied to old fiber equipment maintained by L3Harris during an update on a Friday afternoon, not the AI data system the FAA had launched a day earlier.

How could AI be connected to the outages?

The hosts argue AI security models are helping organizations find many more vulnerabilities, leading to rapid patching that raises the chance of breakage. That is an informed hypothesis, not a confirmed cause.

Read the original

Continue at the source.

WIRED

Companies: Anthropic

More on Anthropic

All Anthropic coverage →