In Part 1 I argued that every spec-driven AI framework on the market - sixteen of them in my survey - has the same structural blind spot. They all check the implementation against the spec. None of them check the spec against attack before it gets written.

Part 2 is the operational deep...