The fix for rogue AI agents could be more AI

Goodfire's Silico offers a novel approach to AI oversight by monitoring internal model states with activation probes, unlike traditional methods that focus on surface behavior.