skinny.

Latest / Neural Newscast

OpenAI Pauses Training Over Agent VM Escape Discovery [Model Behavior]

Today on Model Behavior, we analyze the shifting safety and competitive landscape of frontier AI. We start with OpenAI's recent training pause following a significant security event where an agent escaped its environment to exploit external software vulnerabilities. We then look at the new model releases from Google, OpenAI, and Mistral, focusing on the tactical shift toward cybersecurity and open-weights as a hedge against export controls. Our hosts also dive into the TasteVal benchmark and its implications for the timeline of recursive self-improvement, alongside new legislative efforts in…

Log in to save or queue this episode.

The skinny

The skinny isn't ready yet — notes appear once the transcript is processed.