When the Internet Blinked: What Really Happened During Cloudflare’s November 2025 Outage

Thousands of websites, including ChatGPT and X (Twitter), went down at the same time in the early hours of November 18. A single configuration update gone wrong within Cloudflare’s global network was the cause, not a cyberattack or an overloaded data center.

The Chain Reaction Cloudflare, which acts as a reverse proxy for millions of sites, pushes automated configuration updates worldwide to enforce security policies and threat-mitigation rules. During a sudden surge in traffic, one of these auto-generated rule files grew far larger than expected.

A core control-plane service that disperses security and routing rules to Cloudflare’s edge servers was brought down by that overly large setup. Edge proxies were unable to determine how to forward incoming traffic after that component failed, which resulted in a large number of HTTP 500 errors across websites that relied on Cloudflare’s infrastructure.

Why It Affected ChatGPT and X Even though OpenAI’s and X’s backend servers were healthy, their traffic flows through Cloudflare. When Cloudflare’s routing layer went down, it was like the “front door” to these platforms vanished — requests never reached their origin servers.

How It Was Fixed by Cloudflare Engineers swiftly restored the working version, cleared the cached data on all global nodes, and rolled back to the last stable configuration snapshot. In order to avoid another incorrect push during the implementation of size validation checks, they also momentarily turned off automatic configuration generation. Error rates decreased to baseline within approximately an hour as traffic returned to normal.

As infrastructures rely more on dynamic configuration, even minor control-plane bugs can create global ripple effects. Safeguards like pre-deployment validation, staged rollouts, and sanity limits on generated configs are critical to prevent self-inflicted outages.