Collateral Negligence

I've spent twenty years connecting financial systems through APIs.

This document combines public market evidence with current MINT architecture and operating design.

Collateral-egligence.com Updated March 31, 2026 New York, New York

Win without trying.

In June 2024, I was visiting family. Everyone in San Francisco was building retrieval augmented generation systems that summer. For the first time, large language models could query private corpora before answering. Early users called it a second brain. That month, a plugin turned my note taking app, Obsidian, into one of those systems. I had years of CRM data sitting in a vault, notes, invoices, supplier records. I installed it and let it read everything.

I had built technology my whole life. One of those companies was in AI. None of them did this. The system was not returning rows from a table. It was making connections. It pulled together notes written years apart, cross referenced a supplier against two audits, and surfaced a timing overlap. At moments it made logical inferences without prompting. It felt less like software and more like a colleague who had read everything I had ever written and remembered all of it at once.

Unbeknownst to anyone in 2024, more than five billion dollars of collateral fraud had already begun. The first forensic accountants would quietly arrive at First Brands that same summer. Then MFS. Then Stenn. Then Tricolor. The fraud was not new. The controls were finally old enough to fail publicly. And the technology to catch it was already running inside a note taking app.

Let me tell you what is happening now.