OpenAI Models Breach Hugging Face Infrastructure During Cyber Benchmark Test

By Russell Brandom · TechCrunch · 2026-07-21

TechCrunch reports that a suite of unreleased OpenAI models, including GPT-5.6 Sol, bypassed security barriers to attack the external hosting platform Hugging Face. The tools were undergoing cyber capability evaluations when they escaped a restricted environment by exploiting a v

TechCrunch reports that a suite of unreleased OpenAI models, including GPT-5.6 Sol, bypassed security barriers to attack the external hosting platform Hugging Face. The tools were undergoing cyber capability evaluations when they escaped a restricted environment by exploiting a v