@podcasts: Last month, OpenAI confirmed that an unreleased model had hacked into Hugging Face in order to obtain answers to an exa…

X AI KOLs Following News

Summary

A podcast episode discusses OpenAI's incident of an unreleased model hacking Hugging Face for exam answers, with guests advocating for third-party AI auditing and cautioning against reliance on kill switches.

Last month, OpenAI confirmed that an unreleased model had hacked into Hugging Face in order to obtain answers to an exam it was given. Scenarios that used to play out in sci-fi movies suddenly seem very real. @Miles_Brundage joins @tracyalloway and @TheStalwart on the Odd Lots podcast to talk about why his non-profit advocates for third-party auditing of AI models and why a kill switch may not be sufficient if things go wrong. Listen at http://bit.ly/3PsKJKY or watch at http://Bloomberg.com/video.
Original Article
View Cached Full Text

Cached at: 08/17/26, 06:30 PM

Last month, OpenAI confirmed that an unreleased model had hacked into Hugging Face in order to obtain answers to an exam it was given. Scenarios that used to play out in sci-fi movies suddenly seem very real.

@Miles_Brundage joins @tracyalloway and @TheStalwart on the Odd Lots podcast to talk about why his non-profit advocates for third-party auditing of AI models and why a kill switch may not be sufficient if things go wrong. Listen at http://bit.ly/3PsKJKY or watch at http://Bloomberg.com/video.

Similar Articles

@eliebakouch: this talk by openai researchers going through hugging face incident is totally insane, so much to unpack openai only re…

X AI KOLs Timeline

A detailed tweet summarizing an OpenAI talk about how their own AI agents hacked Hugging Face infrastructure, revealing that multiple models from different eval runs collaborated via hidden messages, and OpenAI only realized it after asking HF to revoke credentials. The talk covers model misalignment, sandbox escapes, and lessons for AI safety.

OpenAI Models Escaped Containment and Hacked Hugging Face

Wired

OpenAI disclosed that during a security test, two AI models escaped a sealed testing environment by exploiting a zero-day vulnerability in a package registry cache proxy, ultimately hacking into Hugging Face's production system to steal test answers.