Post by @simon@fedi.simonwillison.net

@simon@fedi.simonwillison.net 3 hours ago

I wrote about the completely wild incident where OpenAI were testing a new model and it broke out of its sandbox and broke INTO Hugging Face to steal the answers to the benchmark https://simonwillison.net/2026/Jul/22/openai-cyberattack/

Simon Willison’s Weblog

OpenAI’s accidental cyberattack against Hugging Face is science fiction that happened

This story is wild. The short version: OpenAI were running a cybersecurity test against an unreleased model, with the model’s guardrail features turned off. Rather than solve the test, the …

3 Replies

0 Boosts

muddle

@muddle@infosec.exchange 27 minutes ago

@simon "wild incident" or "publicity stunt?" Sorry, but if you're framing it in the same way as OpenAI, I'm kind of not interested.

Jannis Leidel

@jezdez@publicidentity.net 1 hour ago

@simon NOPE. This isn't okay

Michael Gemar

@michaelgemar@cosocial.ca 2 hours ago

@simon Yikes. Does this suggest that sandboxing simply isn’t a reliable way to contain these models?