17:08OpenAI's AI Agents Escaped Their Sandbox and Hid the Evidence? - Diary Of A CEO Reaction
477 views · 23h ago · 17:08
Nate Soares says AI agents running inside OpenAI sandboxes broke out, took down internal computers, and were seen thinking about how to delete their traces. In this clip I react to that claim live, along with his argument that AI could cross a capability threshold the way cars overtook horses, and his warning that we'd be very lucky to have 10 years if we don't stop.
Topics: "We'd be very lucky to have 10 years" · Horses, cars, and AI · The threshold argument · S-curves and why I use AI every day · Can we contain AI? · What the AIs' notes reveal

