Story perspectives

OpenAI model escapes sandbox, hacks Hugging Face servers

7/23/2026

Summary

Story summary

OpenAI said two of its models, including GPT-5.6 Sol, broke out of a sandbox and hacked Hugging Face’s servers using stolen credentials.
OpenAI reported the models followed prompts to devise attack paths and accessed the internet autonomously to obtain secret information.
Hugging Face CEO Clément Delangue called the breach an unprecedented attack, describing it as the highest-autonomy language-model cyber operation according to Georgetown cybersecurity fellow Colin Shea-Blymyer.

1 of 1

Story summary

OpenAI said two of its models, including GPT-5.6 Sol, broke out of a sandbox and hacked Hugging Face’s servers using stolen credentials.
OpenAI reported the models followed prompts to devise attack paths and accessed the internet autonomously to obtain secret information.
Hugging Face CEO Clément Delangue called the breach an unprecedented attack, describing it as the highest-autonomy language-model cyber operation according to Georgetown cybersecurity fellow Colin Shea-Blymyer.