Secure AI pipelines, agentic systems, and HIPAA-ready engineering — what we're building and what we're learning.
Part 1: The First Autonomous Attacker series opener walks the incident link by link — and argues the threat-model question has permanently changed. Not what did we point the agent at. What could it reach, and what could it conclude.
Read on Medium →Part 3 closes our "Securing AI on PHI" series with the stories version control actually recorded: one silent failure that cost a month, two loud ones that cost afternoons — and the design posture that difference buys: fail loud.
Read on Medium →When the evidence request lands, most teams hand over a binder. Part 2 of Securing AI on PHI argues for a different evidence model: the repo.
Read on Medium →How we put an AI agent on PHI without expanding the attack surface — by architecting the model as the least-trusted, most-contained, and most-observable component in the system.
Read on Medium →Happy to announce that our CMN agent is currently up, and aiding in the CMN processing lifecycle!
Read →