A large portion of the internet went dark on Tuesday after a Cloudflare systems failure temporarily knocked major websites—including ChatGPT, X, and Downdetector—offline.

The company has now issued a detailed explanation, confirming the incident was caused by a technical error rather than a cyberattack.

Cloudflare co-founder and CEO Matthew Prince published a blog post Tuesday night outlining what triggered the widespread disruption. The company linked the incident to its Bot Management system—specifically the mechanism that determines which automated crawlers are allowed to scan websites on its network.

Cloudflare, which says 20 percent of global web traffic passes through its infrastructure, uses this system to help websites handle surges in traffic and protect them from DDoS attacks. But on Tuesday, the same safeguards became the cause of a widespread service failure.

How the outage spread across web

The company explained that a permissions change in the database behind its Bot Management engine led to a malfunction. The issue was unrelated to Cloudflare’s newer generative AI safeguards, DNS, or earlier fears of a large-scale cyberattack.

At the centre of the problem was a machine learning model that assigns bot scores to incoming traffic. This model relies on a regularly updated configuration file. However, a recent change to how the system’s ClickHouse queries retrieved data created large numbers of duplicate “feature” rows.

As a result, the configuration file ballooned in size, quickly surpassing designated memory limits. This caused a failure in Cloudflare’s core proxy system, which handles traffic for all customers relying on bot-related rules.

True traffic blocked, false bots allowed

Because of the malfunction, websites using Cloudflare’s bot rules began treating legitimate users as automated bots. This cut off access to countless sites, including ChatGPT and major social platforms.
On the other hand, websites not dependent on these bot scores continued operating normally.

Background

Cloudflare has recently taken a stronger stance on controlling automated crawlers—especially those scraping content to train AI models. Its newly announced “AI Labyrinth” system uses generative AI to overwhelm rule-breaking crawlers.

While that technology was not involved in Tuesday’s outage, the company noted that the failure underscores how heavily the internet relies on centralized infrastructure, similar to recent outages originating from Microsoft Azure and Amazon Web Services.

The company listed four steps it will implement immediately to avoid a repeat incident. While Cloudflare did not detail each step in the report, it acknowledged that increasing dependence on interconnected systems means that some outages may be unavoidable despite stronger safeguards.