Amazon's March 2026 AI outages — two separate incidents within three days, totaling more than 6 million lost orders — have done something unusual for the SRE community: they've made a failure mode visible that most teams have been quietly carrying in their production systems without acknowledging. The incidents were traced to AI-generated code changes deployed without adequate approval gates. Amazon's response was a 90-day code safety reset across 335 critical systems, with a new requirement that AI-assisted code changes be reviewed by senior engineers before deployment. That response is SRE discipline. Applied reactively. This article is about applying it proactively — and about a compounding failure mode most teams haven't modeled yet. The Double-Exposure Problem The SRE concept of blast radius asks: when a component fails, what is the maximum scope of impact? Most blast radius models assume that the failing component is one thing — a service, a database, a network partition.…