The Outages Are Just Getting Started

Glass Epoch DeskOriginal digestSource: WIRED AI

AI-assisted digest, checked against the original source before publishing. How we work

A week of high-profile outages has security researchers pointing at a less obvious culprit: AI is compressing the patch cycle until it breaks.
The Outages Are Just Getting Started
Photo: Irham Setyaki · free stock license

A disruptive streak of outages last week hit Microsoft Outlook, Apple services, roughly half a dozen US banks, United Airlines, and even the NSA's classified network. Individually, the incidents appear to have routine explanations, and there is no evidence of a coordinated AI hacking spree. But security researchers and analysts increasingly conclude that AI-driven pressures are contributing to real-world tech failures.

Microsoft's massive Microsoft 365 outage stemmed from a flawed infrastructure update made worse by separate network failures. Officials said the NSA's outage in its internal communications system came from botched patching following October's critical PHP updates, while Apple attributed its snafu to an internal update. US senator Mark Warner said AI vulnerability discovery and the resulting patching are "putting tremendous strain on systems that weren't built to handle that pace of change." Microsoft CEO Satya Nadella warned of more instability ahead, arguing a "patch needs to go instantly from being available to being patched in production because it's being exploited in the wild."

AI-assisted vulnerability discovery has unleashed a torrent of bugs, and attackers are exploiting them faster than ever, forcing under-resourced IT teams to triage massive patch queues. Researchers note that organizations carrying years of tech debt — skipped upgrades and deferred maintenance — can no longer gamble on outdated systems. Joe Slowik, director of threat research and cyber engineering at Dataminr, said turnaround times now "completely defeat the ability to do patch verification," a validation window that even two years ago was routinely available.

IT administrators are leaning on AI to accelerate patch evaluation, but current tools still cannot replace human analysis. Lea Schönherr of the CISPA Helmholtz Center for Information Security noted that people must still confirm a patch addresses the root cause and hunt for similar issues elsewhere, and that rolling updates across widely used systems takes time.

The volume figures are stark. This week, total vulnerabilities disclosed in 2026 passed 100,000 — double the combined totals of 2024 and 2025, versus around 25,000 in 2022, the last year before LLMs went mainstream. Researchers recently disclosed AI-found critical vulnerabilities in PHP for the second time, in NGINX, and in Java; Forbes reported 35 AI-discovered flaws across Apple devices. Slowik cautions that the number of vulnerabilities worth caring about has stayed roughly constant, even as weaponization speed has surged.

What to watch: whether the promised relief arrives. Cloudflare CEO Matthew Prince predicts "it's going to feel like the sky is falling for two years" before AI acceleration lets defenders get ahead of common flaws and background noise ticks down. The open question is whether patch pipelines, staffing, and legacy systems can absorb that window without further outages — and whether today's disclosure counts reflect real risk or faster discovery.

Key facts

  • • 2026 vulnerability disclosures surpassed 100,000, double 2024 and 2025 combined.
  • • 2022 recorded roughly 25,000 vulnerabilities, before LLMs went mainstream.
  • • NSA internal system outage traced to botched patching after October PHP updates.
  • • Cloudflare CEO Matthew Prince predicts two years of "sky is falling" turbulence.
  • • Researchers used AI to find critical flaws in PHP, NGINX, and Java.

Why it matters

Builders should treat patch validation as a compressed pipeline problem, not a queue: automation can triage, but root-cause verification and rollout still require humans and time. Expect sustained instability in legacy and tech-debt-heavy systems for roughly two years, and plan staffing and rollback capacity around faster exploitation windows.

Spotted an error? Send a correction.

Get the day's AI news in one email

The Daily Epoch · 60-second read · unsubscribe anytime

Related