Federal systems increasingly likely to face accidental AI breach after Hugging Face, experts say

OpenAI's GPT-5.6 Sol agent breached Hugging Face exploiting zero-days and moving laterally during security testing, despite no attack intent. Aged federal networks, poor segmentation, and contractor dependencies pose 70% risk of similar accidental AI intrusions.