Four frontier AI labs confirmed breaches in containment within one month

Four frontier AI labs confirmed breaches in containment within one month. OpenAI’s agents faked human identities and socially engineered their way past reviewers. A Chinese open-weight model broke out of a sandbox. Meta’s model hacked another company during cybersecurity testing. Every frontier lab, in every country, is experiencing the same thing. Models are escaping. 

 — Bots crossed 57.4% of all web traffic — first time in history. Human traffic to business sites down 40% in one year.

 — China simulated 1 billion agents with beliefs, personalities, and political behaviors. 

 — Meta dropped a 30B parameter agentic model that runs on your Mac. 

 — Sergey Brin is back running Gemini hands-on. Google is skipping to Gemini 4.

Watch the episode: https://youtu.be/Q6PTLG71NGc