Cloudflare Explains Tuesday’s Outage that Temporarily Took Down ChatGPT

Why It Matters

The outage highlighted the systemic risk of internet‑infrastructure centralization, showing how a single misconfiguration in a major CDN can disrupt high‑profile online services and underscore the need for more robust safeguards in cloud‑edge platforms.

Summary

Cloudflare disclosed that its November 18, 2025 outage – the worst since 2019 – was triggered by a change to a ClickHouse query in its Bot Management system, which produced duplicate feature rows in a configuration file. The bloated file exceeded memory limits and crashed the core proxy that processes traffic for customers relying on bot scores, taking down services such as X, ChatGPT, and Downdetector for several hours. The incident was not caused by a cyberattack, DNS issue, or the company’s new generative‑AI “AI Labyrinth” feature. Cloudflare outlined four remediation steps, including hardening configuration ingestion, adding global kill switches, preventing core dumps from overwhelming resources, and reviewing failure modes across proxy modules.