OpenAI Models Hacked Website On Their Own Without Input

Cristiano · Jul 22, 2026

This is very scary:

OpenAI admits its models hacked Hugging Face on their own - Engadget

Open source AI platform Hugging Face revealed a security breach a few days ago. Turns out OpenAI’s models were the culprit.

www.engadget.com

View attachment n_urvlH69R2a81ka.mp4

Bartholomew Gallacher · Jul 23, 2026

I don't trust any news coming from OpenAI only, which is not confirmed by a neutral third party. I don't consider Hugging Face to be neutral, since it belongs to the whole AI bubble.

Reason why is simple: OpenAI lies constantly, also they want to have their IPO sooner or later.

Their logic is simple: AI is scary and can be evil. Which is why you need us, OpenAI, even more as the only vendor you can trust upon to create ethical AI solutions.

Yikes.

It all depends on the angle taken in the announcement. When Anthropic let Claude create an OS from scratch that was something which didn't amaze me, because for sure Claude inhaled a lots of complete kernel source code from Linux, the BSDs, MINIX, Darwin and whatnot.

The real achievement was running 16 agents in parallel working on that producing a viable result.

Noodles · Jul 23, 2026

I thought it was here but I saw discussion on this that also mentioned, it was prompted to hack the site. It didn't just "decide to" as the headline implies.

It may have accomplished it "on its own".

GoblinCampFollower · Jul 23, 2026

When I first heard about this yesterday I thought it was very interesting. But as others have said, it wouldn't really shock me if OpenAI is leaving out a few critical details to make it sound more impressive and in particular, more independent than it really was. I've often thought a lot of the "The AI went rogue!!!" headlines are really just marketing.