The Day the Internet “Died”: What Really Happened When ChatGPT Went Down
Do you remember when you all opened OpenAI’s Chatgpt, X (formerly Twitter), Perplexity and other platforms displaying an error message telling us: “please unblock challenges cloudflare.com to proceed.”
For a moment, the entire digital world felt like it hit the pause button. People joked, “If ChatGPT is down… are we supposed to think for ourselves now?”
But behind the memes, it was a major tech glitch. Let’s understand what actually happened.
OpenAI’s systems are massive, think of a bustling city where every building is connected through invisible skybridges. If one building collapses, the whole city feels the shock.
The main disruption was caused from a major Cloudfare outage, not just OpenAI. Cloudfare’s outage caused widespread 500 errors😵
What is Cloudflare?
Cloudflare is one of the biggest networks operating company and is used by many websites to keep speed up their service and protect them from online attacks. The platform acts as a middle layer between websites and the people trying to access them.
Cloudflare provides critical infrastructure (CDN + DNS + security) for a huge number of websites — roughly 20% of the internet uses its services.
It is this company that works quietly behind the scenes to keep internet running smooth. Most of the time, their work goes unnoticed, but when something goes wrong, their absence becomes instantly visible. Since Cloudflare supports a number of companies, a technical issue on its end can lead to outages on multiple unrelated platforms.
How the outage started?
Cloudfare first posted that it was aware of an issue around 11:48 UTC. It noted some internal server errors and instability in some regions. So while working on a fix, Cloudfare disabled WARP access in London.
What caused this error?
Cloudfare maintains a file to detect and filter out malicious or automated traffic. It is called “Feature file” under Bot Management System. Due to changes in a database (Clickhouse) permissions, a query started generating duplicate entries in this feature file. Due to this, the file size became so large, that it caused the software to crash. This crash disrupted many of it’s core services, which led to errors.
Get Payal Makwana’s stories in your inbox
Join Medium for free to get updates from this writer.
At first, Cloudflare thought the symptoms might be caused by a massive DDoS attack because of the spike in “unusual traffic.”
But after detailed investigation, they found the real cause was internal, the corrupted/ large configuration file, that led to widespread 5xxx server errors.
The Fix and Rollback
Errors continued until the underlying issue was identified and resolved starting at 14:30. Cloudfare stopped the generation and propagation of the bad feature file and manually inserting a known good file into the feature file distribution queue. And then forcing a restart of core proxy services to restore the normal traffic flow. Everything was normal by ~17:06 UTC, All downstream services restarted and all operations fully restored.
Cloudfare officially aplogized on behalf of their entire team.
A Strange Lesson
Millions of people realized something that day:
We rely on AI more than we think.
From writing messages…
to generating ideas…
to debugging code…
to making presentations…
to learning new skills…
AI has quietly become the electricity of the digital world.
Sometimes, the value isn’t noticed until it’s absence is felt.
And when it was gone, even for a few hours, it exposed how deeply it has embedded itself in our daily lives.
The outage did not indicate a lack of strength.
It served as a reminder that we are creating something extraordinarily intricate and human.
