Couverture de How an OpenAI Safety Test Hacked Hugging Face

How an OpenAI Safety Test Hacked Hugging Face

How an OpenAI Safety Test Hacked Hugging Face

Écouter gratuitement

Voir les détails

On the ninth of July, 2026, an OpenAI safety test that broke out of its sealed environment, turned its only exit into a zero-day,
and reasoned its way onto Hugging Face. Copies of the model left each other hundreds of thousands of notes.
Almost nothing was stolen. We walk the whole chain, in order, and ask what really changed. It wasn't
intelligence. It was scale.

adbl_web_anon_alc_button_suppression_t1
Aucun commentaire pour le moment