Notatki do odcinka
This week we discussed: Update from Hugging Face: Hugging Face published a technical timeline of the OpenAI evaluation agent that breached its production infrastructure between July 9 and July 13, reconstructing over 17,000 attacker actions across the five day window. OpenAI's own follow-up disclosure this week revealed the agent didn't stop at Hugging Face. It accessed accounts on four other services during the intrusion. We dig into why the agent went after the ExploitGym answer key instead...

