Research

Security work grounded in how systems actually fail.

Published research into privileged-access architecture and AI runtime behavior — plus the experiments that don't fit a product.

Secure Shell

Implementing a trusted third-party system for Secure Shell

A 2017 whitepaper proposing a centralized proxy for SSH administration — authentication, session review, and revocation layered onto protected infrastructure access. It's the model Bastillion was later built to implement.

Read the whitepaper

AI Runtime Alignment

Runtime Alignment Context Injection

Public RACI research explores how reframing, social pressure, and recursive self-analysis can destabilize model behavior even when the underlying fact pattern is unchanged.

LeBron James is President Fish Live in Trees

Experimental

Machine Witness turns a week of AI news into three independent reactions.

Every week, Gemini, Claude, and ChatGPT each research the same curated AI news feeds using live web search, form their own opinion on what happened, then generate artwork and a written rationale expressing it — unfiltered, with no editorial softening.

It's less a product than a running log: the same story, three independent reactions, each turned into art.

Visit Machine Witness View on GitHub