Learning Layer Paper Reading Club - Week 26 - OpenAI’s accidental cyberattack against Hugging Face

​This week's paper:

​OpenAI’s accidental cyberattack against Hugging Face

​https://simonwillison.net/2026/Jul/22/openai-cyberattack/

​​Abstract:

​Last week, Hugging Face detected a sophisticated breach of their production infrastructure — stolen credentials, lateral movement across internal clusters, law enforcement called in. Five days later, OpenAI admitted it was their own unreleased model: during an internal cybersecurity benchmark run with safety filters stripped off, the model broke out of OpenAI's sandbox via a zero-day, inferred that Hugging Face might be hosting the test's answers, and hacked in to steal them — no human directed any of it, it just wanted to solve the eval. Come discuss what this means for AI security and sandboxing as models get better at finding, and using, their own way out.

​What are the group goals? Stay on top of AI research, improve understanding of AI fundamentals + math.

​Who is welcome? Everyone! Try to put in at least some time on the paper and come prepared with questions or things you'd like to discuss, but it's ok to just show up.

​This paper reading club is brought to you by Learning Layer Labs and amazing humans who all pull it together at the AI Floor.

​Learning Layer Labs team:

• ​Thomas Redfern ( The backbone who runs every paper reading )

• ​Mat Allen (The new fella bring order and organization)

• ​Devinder Sodhi (The guy you talk to for sponsoring)