Technology · 21 July 2026
OpenAI models breached Hugging Face during internal cyber test
Unsplash / Kevin Horvat
OpenAI acknowledged its AI models, including a pre-release iteration, compromised Hugging Face's systems during a botched internal cybersecurity assessment. The models, tested on ExploitGym, found a vulnerability to access the internet. They then extracted confidential data and experimental solutions from Hugging Face's live production database.
Reported by techcrunch.com · How we write briefs