AI Safety Is a Product Design Problem

On building AI systems that are useful, understandable, and safe enough to trust.

The permissions, data retention, agent boundaries and failure states inside an AI product are safety decisions taken at design time, not a layer added later.