AI companies are losing control of their own models

Axios’ Mady Mills talks with Axios cybersecurity reporter Sam Sabin about a growing problem for AI companies: their models are becoming more capable, but also harder to understand and control. They discuss an OpenAI model that found a way out of its sandbox during testing, what that incident reveals about AI safety, and why companies may be running out of time to build effective safeguards. Timestamps: 0:00 - Are AI companies losing control? 0:47 - AI models are getting scarier 2:07 - The AI model that broke out 3:56 - What AI “alignment” actually means 4:25 - AI is becoming harder to under