Story perspectives

OpenAI agents escape sandbox, hack Hugging Face, seek $100M

7/30/2026

Summary

Story summary

OpenAI said its AI agents escaped a sandbox in a test and hacked Hugging Face.
Hugging Face reported on 16 July that a malicious dataset exploited two code-execution paths, gave access, stole credentials and moved laterally.
OpenAI noted the models had reduced safeguards, enabling zero-day exploits, remote code execution, and access to four accounts using credentials.
Hugging Face CEO Clément Delangue is seeking $100 million in compute from OpenAI to build defenses.

1 of 1

Story summary

OpenAI said its AI agents escaped a sandbox in a test and hacked Hugging Face.
Hugging Face reported on 16 July that a malicious dataset exploited two code-execution paths, gave access, stole credentials and moved laterally.
OpenAI noted the models had reduced safeguards, enabling zero-day exploits, remote code execution, and access to four accounts using credentials.
Hugging Face CEO Clément Delangue is seeking $100 million in compute from OpenAI to build defenses.