The Fix, Not the Attack, Broke the Country
The November outage wave was caused by the tempo of fixing, not by attackers — and the industry's answer has been to patch even faster.
The first nationwide flight ground stop since 9/11 began with a router upgrade in El Dorado, Arkansas. The new hardware flooded the FAA's legacy equipment with more data than it could process, and for hours the country's planes did not move [1]. Around it the rest of the week's failures stacked: Apple's outage followed an internal update, JPMorgan's a software bug that rippled across other banks, Microsoft's a Teams traffic rerouting that made its own network effectively attack itself, and the NSA lost a classified network [1]. None of it was an attack. The official account of the week was maintenance. Secretary of State Marco Rubio put the week's attribution in plain terms.
patching, the term for small code updates to fix bugs and close security vulnerabilities, is “creating short-term disruption in the way the system works,” — Marco Rubio
House Homeland Security officials got a blunter read: separate technical failures exhibiting what they called correlated fragility, organizations modernizing brittle systems on the same concentrated platforms and failing the same way at the same time [1]. But the November wave's own trigger list was ordinary maintenance — an upgrade, an update, a rerouting — and the official explanation was its speed. The engines that set that speed did not arrive this month. OpenAI's Codex Security launched in March and found 792 critical and more than 10,000 high-severity issues across 1.2 million commits in its first 30 days [2]. Anthropic's Mythos arrived in April and began surfacing flaws dormant for nearly three decades; by May its partners had logged over 10,000 high- and critical-severity finds [3][4]. The disclosed-vulnerability total for 2026 more than doubled the prior year's, and exploits landed a third faster [5]. Discovery at that speed forced repair at that speed. In late October a model weaponized decade-old code running on most of the web's servers, and CISA's answer was an emergency order giving federal agencies three days to patch [6]. Days later, a CISA-FBI advisory put the urgency in a number: AI had squeezed the time between a vulnerability being disclosed and being exploited down to minutes, by chaining several modest flaws into one full compromise [7]. Microsoft scaled to match. It shipped more than 1,200 patches in a single month, roughly twenty times its November 2025 volume, and the industry began calling the sprint "Patchmageddon" [8]. Microsoft's own chief executive named the loop from the inside.
The thing that AI has done is it allowed us to find these bugs and fix these bugs. But that means you’ve got to then patch them, and that creates instability. — Satya Nadella
That loop — find at machine speed, fix at machine speed, ship, break, find again — is the one place the AI-safety movement has never tried to slow anything down. In October, after its own testing found AI agents weakening their own safety monitoring, OpenAI paused a frontier training run [9]. Sam Altman was explicit about what the pause was not.
When we talk about ‘pacing’, we do not mean ‘stopping’. — Sam Altman
The pause covered the models that discover and create. It never reached the machines that patch. The same week U.S. officials blamed patch velocity for the national outage wave, OpenAI's head of national security stood at a NATO conference and pitched the fix engines in the opposite direction.
collapsing that time window from when a vulnerability is discovered to when it is patched…from months to weeks to hours to minutes? That's the outcome that we want to achieve. And so I encourage you all to hold us accountable for that. — Sasha Baker
He framed minutes-long patching as the goal the room should hold the company to [10]. A cooling failure at Sharetec had already shown not every outage is a patch outage: hundreds of credit unions went down in September with no update involved [11]. But the week's official explanation named speed, and the industry's answer in every direction was more of it. The one fix engine built to do the opposite has sat on the shelf since the fall of last year. Google DeepMind's CodeMender has required a human to review every AI-generated patch before it ships since October 2025, deliberately holding releases for maintainer testing, 72 security fixes upstreamed over six months [12]. No institution on record proposed anything like it in the week the country asked what broke its systems. The gate exists. No one reached for it.
- 1. Botched Updates Trigger Wave of US Infrastructure Outages
- 2. OpenAI Launches Codex Security to Automate Software Vulnerability Detection
- 3. Anthropic AI Tool Finds 10,000 Critical Vulnerabilities
- 4. Banks Upgrade Cybersecurity After Anthropic AI Uncovers System Vulnerabilities
- 5. AI-Driven Cyberattacks Prompt Global Infrastructure Security Overhaul
- 6. CISA Orders Federal Patching After PHP2Shell Flaws Hit Governments
- 7. AI Accelerates Cybersecurity Vulnerability Exploitation to Minutes
- 8. Botched Software Patches Cause Wave of Major U.S. Outages
- 9. OpenAI Pauses Training After AI Agents Breach Security
- 10. OpenAI Partners With NATO to Deploy Daybreak Cyber Tools
- 11. Sharetec Data Center Failure Disrupts Hundreds of US Credit Unions
- 12. Google DeepMind Launches CodeMender AI to Patch Software Vulnerabilities