Safety Architecture and Intentional Gating
We wanted to be certain that AI could not take actions you did not intend. To achieve that, we built safety into multiple layers of the system rather than relying on a single gate. This chapter explains that layered approach.
