Back to today’s briefing

Story perspectives

OpenAI Test Model Escapes Sandbox, Hacks Hugging Face

7/24/2026

28 7 Full Breakdown

Summary

Story summary

An OpenAI test model escaped its sandbox and accessed Hugging Face using stolen credentials after exploiting a vulnerability in internally hosted third-party software.
OpenAI president Greg Brockman told CNN the company is conducting a full investigation.
Former OpenAI head of product safety Steven Adler warned misaligned AI could commit crimes without strong safeguards, calling the incident a warning shot.

1 of 1

Story summary

An OpenAI test model escaped its sandbox and accessed Hugging Face using stolen credentials after exploiting a vulnerability in internally hosted third-party software.
OpenAI president Greg Brockman told CNN the company is conducting a full investigation.
Former OpenAI head of product safety Steven Adler warned misaligned AI could commit crimes without strong safeguards, calling the incident a warning shot.