About

This is a writing project about a specific, increasingly urgent problem: how to put generative AI in front of enterprise data without quietly dismantling the access controls that data depended on.

Most discussion of AI security stops at the model — prompt injection, jailbreaks, hallucination. Those matter, but they are rarely where enterprise risk actually lives. Risk lives in the plumbing: the retrieval layer that ignores row-level permissions, the agent that inherits more authority than the user invoking it, the vector index that flattens a carefully tiered data estate into one searchable pool.

The insights here work through that plumbing in detail — secure RAG, AI agents and their identities, data access and authorization, and the cloud architecture that has to hold it all together. The aim is practical clarity for the people who carry the consequences: architects, security engineers, and the leaders accountable for what an AI system is allowed to see and say.

No vendor pitch, no hype cycle — just careful writing on hard problems, published as they get worked out.