Episodic Memory at the Gateway Seam
Write-time extraction proposed 700 entries a day and nobody reviewed them. The replacement: owner-private, PII-sanitized episodes, at most eight cited lessons a day, and a review ceremony at the share gate.
Read more →Pending by Default: Governing What Your AI Remembers
Memory amplifies — one write becomes many reads. Four design choices that put a human between capture and recall: pending-by-default review, durable rejection, scoped reads, and idempotent injection.
Read more →An Enforceable Framework for Operating LLMs Safely
Seven pillars for running LLMs inside an organization — and the test that separates a real control from a checkbox: on every path, verifiably, or not at all.
Read more →A Cortex Can't Coordinate
An LLM is the creative cortex. Cortices don't coordinate. When three agents touch the same file, the substrate underneath has to. Here's what that looks like.
Read more →Shadow AI: The Blind Spot Application Gateways Miss
Application-level gateways only see traffic you explicitly route through them. Network-level visibility catches the AI use they can't — Claude CLI, browser ChatGPT, personal-device apps.
Read more →Why AI Gateways Aren't Enough
Security platforms monitor AI. Routing gateways proxy AI. DLP tools scan AI. But they're all isolated. The integration gap is where risk lives.
Read more →What a Brain Can Teach Us About AI Security
LLMs are the creative cortex. But a real brain needs memory, threat detection, judgment, and routing. Here's what that looks like as infrastructure.
Read more →