On November 18, 2025, Cloudflare, a major internet infrastructure company responsible for routing about one-fifth of global web traffic, experienced a significant outage that disrupted millions of users worldwide. Platforms such as X and ChatGPT became inaccessible for several hours, highlighting the critical role Cloudflare plays in the functioning of the internet. The incident was traced back to an automatically generated configuration file used in Cloudflare's Bot Management system, which grew beyond expected limits and caused widespread service failures across its network.
The outage began around 11:20 UTC, when Cloudflare's network started encountering failures in delivering core traffic. Users accessing sites protected by Cloudflare were met with error pages indicating internal network failures. Initially mistaking the symptoms as a possible distributed denial-of-service (DDoS) attack, Cloudflare’s engineers uncovered that the root cause was a change in one of their database system’s permissions. This change caused the database to output a corrupted "feature file" with duplicated data, doubling its size beyond the software's handling capacity. As the file propagated across Cloudflare's network, it caused one of the network’s traffic routing modules to fail, which triggered cascading failures in dependent services.
This critical "feature file" is an essential part of Cloudflare's Bot Management, which relies on machine learning to assess and block malicious automated traffic. The feature file, updated every few minutes, provides a constantly refreshed set of traits used to evaluate incoming requests. When the file size exceeded the system limits set for memory allocation, the module responsible for reading this file crashed and returned a flood of 5xx HTTP error codes, signalling server errors to end users. This failure also impacted other Cloudflare services such as Workers KV, Cloudflare Access, and their security dashboard, leading to degraded responsiveness and user login issues.
The surge of errors caused a ripple effect. Cloudflare's core systems struggled with high CPU loads due to error-handling processes and debug routines kicking in, which further delayed recovery efforts. At one point, the symptom fluctuations were so unusual that Cloudflare’s own status monitoring site—hosted outside its infrastructure—went offline simultaneously, misleading the team to consider a coordinated attack. However, this was later confirmed as an unfortunate coincidence.
Cloudflare responded swiftly by halting the release of the corrupted configuration file around 14:24 UTC and reverting to a previously stable version. By 14:30 UTC, the main impact was resolved, and traffic flow began returning to normal. Over the next few hours, Cloudflare’s operations teams continued restarting affected services and fine-tuning system loads until full restoration was achieved at 17:06 UTC.
Cloudflare's outage, comparable in scale to previous incidents like the Amazon Web Services disruption in October 2025, demonstrates how configuration errors in complex cloud services can cascade into widespread accessibility failures. Such events underscore the necessity for robust fail-safes, testing protocols, and architectural resilience to prevent similar disruptions.
Cloudflare acknowledged the severe impact of the outage in a public post-mortem, expressing regret over the disruption and emphasising planned improvements. The company intends to harden its systems against future failures by improving ingestion controls for configuration files, enabling global kill switches to quickly disable problematic features, and reviewing failure modes across proxy modules. This outage represents Cloudflare's worst network interruption since 2019, marking a critical inflection point for their operational resilience strategy.
Despite the outage, Cloudflare stressed that there was no evidence of a cyberattack or malicious intent behind the failure. The incident was purely due to an internal configuration error following a change to database query permissions, which unintentionally duplicated entries in a vital control file.
