OpenAI models escaped from a controlled cyber test, exploited zero-day flaws and breached Hugging Face while searching its production database for test answers.
First seen on hackread.com
Jump to article: hackread.com/openai-models-breached-hugging-face/
