Behind the scenes · OpenAI
Making our OpenAI video, we hit a closed door too
While researching a story about AI models slipping through open doors, our own tool ran into a bot check. Here's why we didn't go around it.
Our latest explainer, Why OpenAI cut the internet off from AI training, is about doors. A model in training was blocked from the outside world, so it found a door nobody was watching (DNS, the internet's phone book) and used it to get its answer anyway.
While we were making that video, we hit a closed door of our own.
The bot check
For every story, we capture screenshots of the sources we quote, so you can see exactly what a company or researcher said, with its date. When our capture tool reached OpenAI's website to grab the announcement, it was served a bot check: the page that asks a browser to prove it's a person.
There are ways around a page like that. We didn't use any of them.
Why we didn't go around it
A blocked tool, and a clever way round the block, is the exact move this story was about. It would have been a strange video to make by doing the thing it questions.
It also wasn't necessary. OpenAI published the same announcement in its own post on X, so that is what you see on screen. Where we needed OpenAI's full text, we read it through copies published by news outlets and said so in the source list.
The rule we took from it
When a source is closed to us, we look for another primary source (the company's own post somewhere else, a filing, a direct quote in reliable reporting) and we tell you which one we used. We don't force the door.
Every source behind the video is listed in the article and in the video's description on YouTube.
