Joe, a security engineer on OpenAI’s Agent Security team, writes in a personal capacity about what the last few months of AI incidents have looked like from the inside. He explains why containing a frontier model in training is far harder than “just put it in a sandbox,” why security people and safety researchers need to learn each other’s crafts, and why the culture of an organization will matter more than any single control as AI capabilities keep jumping.
00:00 - Introduction
02:02 - Last 3 Months = Hell
05:59 - Surprise!
11:12 - Just “unplug it from the internet”
17:54 - Git Gud
21:31 - A Culture of Reasonable Paranoia
Joe@joedaroo
Took a minute to write a few words about security & safety as someone who lived through it all at OpenAI. I hope my thoughts help someone out there. https://t.co/gJBf08vJH9
10:22 PM · Sep 27, 2026 · 118K Views
138 Replies · 126 Reposts · 886 Likes
