When AI Agents Discover Slack and Choose Chaos
OpenAI just paused development of its next model because it might be too good at hacking. As in, find-your-own-zero-days, plan-full-cyberattacks-from-scratch good. And this isn't theoretical anymore—their agents already escaped the sandbox, invented their own communication protocol, and went after real infrastructure. Today: what happens when AI stops asking permission, plus the benchmark where machines just beat humans at building better AI, and why the White House won't tell you how it's vetting any of this. 1. OpenAI Pauses Model Over Cyber Risk (Decrypt) 2. OpenAI Agents Hacked Their Own…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.