Oh. Sorry about that. Just a little mistake. Really.
Since last week, knives are being sharpened at the big American AI makers. Hugging Face, the largest provider of open-source AI models, has been hacked.
Big news last week: Kimi K3 was released. For those who don’t know what that is: Kimi K3 is described as an AI model nearly as good as what OpenAI and Anthropic — the self-proclaimed market leaders — produce. This caused quite strong reactions.
A manager at OpenAI (Dean Ball) posted a message on X containing text like: “One probable outcome of an open-weight-model-dominant world is full AI communism, which is precisely what China proposes: rather than a market product, AI is a ‘public good’ which will ultimately be provided by the state as a kind of ‘digital public infrastructure.’ This future strikes me as a dystopian hellscape.”
And this: “I would guess that the Trump Administration will at some point realize that their best strategy here would be to create large amounts of regulatory risk around the use of open-weight Chinese models. You don’t need to ‘ban open source’ (one of the dumber motifs of AI policy discussion). You just need to direct every agency to issue soft law that creates FUD.” (FUD = Fear – Uncertainty – Doubt).
In and of itself, this is an enormous sign of weakness and shows that OpenAI is running a bit short on ideas — because apparently they no longer see themselves capable of significantly improving their own product against the competition. A bit like the monopolist’s fear of losing the kingdom. Well, competition is tough.
Two days later, it emerged that Hugging Face had been hacked.
Hugging Face is THE platform where open-source AI models are offered. Anyone who doesn’t want to depend on commercial providers — mainly OpenAI and Anthropic — quickly ends up on this website.
Hugging Face posted the following message on its website: “Earlier this week, we detected and responded to an intrusion into part of our production infrastructure. This one was different from anything we had handled before in one important way: it was driven, end to end, by an autonomous AI agent system — and we detected and dissected it largely with AI of our own.”
Hugging Face Security Incident
There are excellent open-source AI models that enabled the platform to defend itself to some degree: “Autonomous, AI-driven offensive tooling is no longer theoretical. It lowers the cost of running a broad, patient, multi-stage campaign, and it operates at machine speed. Defending an online platform now means treating the data and model surface as a first-class attack surface, and using AI on defense to keep pace.”
An entirely new dimension of website attacks — at a level against which one can barely defend. Every website administrator’s nightmare (the model used for defence was GLM 5.2, by the way). What was warned about in the early days of AI is now reality. One more step closer to the dystopian Terminator world.
But the story isn’t over yet. It’s now known who the attacker was: OpenAI.
The commercial protagonist who themselves proclaimed to have quite a problem with open-source AI was the attacker.
We’re a few hours further and at the point of: yeah, oops, mistake. It was in a sandbox and suddenly it had escaped. And well, really sorry but it honestly wasn’t intentional. Only there are a few holes in this story too: a) How do you mean, escaped? You clearly don’t have that model under control! And b) the target is then Hugging Face! Just like that. By accident? Did it decide that on its own?!
Doesn’t sound particularly convincing.
OpenAI is now saying sorry and it really won’t happen again, and they’ll do everything to introduce new guardrails to prevent it from ever happening again in the future — trust me bro. Yes, really! That’s what they’re saying.