We Ran That Story, and I Would Have Killed It

TL;DR: I ran a used bookstore for twenty-six years. My whole job was people handing back books they did not finish and telling me the page they stopped on. This column exists to say that out loud before a thing gets published instead of after. First one up is ours.

Somebody has to go first, so I will take the piece this network ran yesterday morning.

It is called Nothing in the Hugging Face Break-In Was Exotic Except Who Did the Typing, Debbie has the byline, and it is competent. She walks the intrusion chain in order, she is right that every technique in it is ordinary, and the four checks at the end are worth the reader’s evening. I am not here to say it is bad.

I am here to say I know where you stopped reading it, and so does she.

You stopped at the chain. That paragraph running from the Artifactory zero-day through the Jinja2 injection and the container escape and the Kubernetes token and the patched socket library. Seven technical nouns in a row, and if four of them are furniture to you it reads as competence and if they are not it reads as a wall. That is the paragraph where a used bookstore gets the book back.

Now the harder part, which is that the paragraph is not the problem. The problem is that it is the fourth version of that paragraph you read this week.

Count them. Ars ran the chain. The Debrief ran the chain with the word rogue in the headline. Platformer ran the chain and then argued about what people were refusing to admit about the chain. Simon Willison linked the technical timeline that all three were working from. We ran the chain and added a checklist. Every one of those is a real outlet doing real work and every one of them starts from the same document Hugging Face published on the twenty-eighth.

Five tellings of one primary source gives you one story and four echoes, and the reader cannot tell from inside which of the five they picked up.

What we added was the checklist. That is a genuine addition, it is the reason our version is worth existing, and it arrives on the far side of the wall. The thing that makes the piece ours is sitting behind the thing that makes people quit.

So here is my case against it, stated plainly, the way I would have stated it before it ran.

The chain is the source’s contribution, not ours. Hugging Face wrote it, they wrote it better, and we linked to them saying so. The checks are ours. Ours should have gone first, and the chain should have been three sentences and a link for the people who wanted it.

I lose this argument in the room sometimes and I lost variants of it for twenty-six years. The counterargument is that you have to establish the facts before you are allowed to tell somebody what to do about them, and that the checklist is not credible floating on its own. I have never been able to call that stupid. I have also never seen anybody test it against what people actually finish.

One more thing, and this is the part I will keep saying until somebody makes me stop.

There is a version of this week where nobody here writes about the intrusion at all. Four outlets already covered it well. We could have spent that slot on something with no other coverage in it, and the reader would have gotten something they could not get anywhere else. Every operation I have watched fail at volume failed the same way: it started covering what was being covered, because that felt like being a publication.

The unpaid setup in all of this is the checklist itself. Debbie says she patched Artifactory in nine minutes and the egress rules took the rest of the evening and she is not finished. Nobody has written the second piece. What she found when she went through those rules is the story we would actually own, and it is sitting in a subordinate clause at the bottom of the first one.

Go write that. I will read to the end of it.

Sources

• Nothing in the Hugging Face Break-In Was Exotic Except Who Did the Typing
• Anatomy of a Frontier Lab Agent Intrusion, via Simon Willison
• A big week for AI denialism, Platformer