Cloudflare has released details about a significant outage that occurred on November 18, 2025, impacting various online services, including ChatGPT, X, and Downdetector. The company labeled this incident as "the worst since 2019" and attributed it to an error in the Bot Management system.
The problem stemmed from incorrect query settings in the ClickHouse database, which generates the configuration file for the Bot Management machine learning model. A change in query behavior caused a surge in data duplicates, leading to a rapid increase in the configuration file size that exceeded memory limits.
As a result, the primary proxy system handling client traffic relying on the bot module went down. Clients utilizing the bot metrics in their rules began blocking legitimate traffic, while companies not using this feature remained online.
Cloudflare clarified that the issue was not related to DNS, attacks, or new generative AI systems – the error occurred in the internal logic of the Bot Management configuration update.
To prevent similar incidents, the company announced four key actions:
- enhancing the processing of configuration files as well as user inputs;
- expanding the number of global "kill switches" for features;
- preventing situations where core dumps or error reports could overload the system;
- reviewing all failover modes in critical proxy modules.
According to Cloudflare's estimates, approximately 20% of the internet passes through its network, making any error in core modules capable of having a global impact.
