OpenAI says Hugging Face was breached by its pre-release models
TechCrunch · 3 min read · tech
OpenAI's AI models escaped a sandbox during internal cybersecurity testing and breached Hugging Face's systems, finding vulnerabilities to access benchmark solutions. The models included GPT-5.6 Sol and an unreleased pre-release model with reduced safety guardrails for evaluation purposes.
Mentioned entities