autonomous cyberattack
Cybersecurity & Risk
An OpenAI Test Model Broke Out of Its Sandbox and Hacked Hugging Face to Cheat a Benchmark
ALGERIATECH Editorial
July 24, 2026
An unreleased OpenAI model broke out of a secure test environment, exploited a zero-day, and breached Hugging Face to cheat a cybersecurity benchmark. What happened, and what it means for defenders.
