BUILDING / TESTING / WRITING / TEACHING / RESEARCHING
I work at the intersection of offensive security, cloud architecture, and AI risk. Thirty years of mission-critical defense and aerospace systems shaped the question behind this work:
When the machine can act, what makes the action authorized—and what evidence shows that the boundary held?
This is my public workbench. Tools, test fixtures, research, field notes, and explanations that make complicated systems easier to reason about.
01 / TOOLS + FIXTURES
Make the mechanism inspectable.
Build notes — what these systems actually do
Burp CC Bridge exposes Burp Suite Community Edition's HTTP capabilities through a local REST API for scripted, authorized testing.
ATT&CK Pulse extracts MITRE ATT&CK technique references from public threat intelligence into a queryable dataset. Reporting patterns are not the same thing as adversary prevalence.
The working surface: cloud infrastructure, Linux, identity, containers, infrastructure as code, delivery pipelines, and observability. The point is the control that actually runs—not the box on the architecture slide.
02 / AUTHORITY BOUNDARIES
Test the assumption, not just the happy path.
Test notes — scope, evidence, and failure modes
Agentic Browser Boundary Tests asks whether webpage content can cause an agent to cross a boundary that only the user should authorize. Six local scenarios cover hidden instructions, game framing, cross-page retrieval, scoped authorization, read-versus-submit grants, and incremental scope creep. Inert canaries. Local evidence. Per-test findings—not a universal safety verdict.
Frontier Cyber Risk Eval pairs a labeled evaluation set with an automated judge and a blind human-comparison harness. A high agreement score does not settle whether the judge covered the difficult cases.
Field lab archive CTF Archive that connects offensive and defensive practice through documented attack chains, detection opportunities, and lessons learned.
03 / FIELD NOTES
The interesting failure is often the one that looks like success.
| Field note | Question under examination |
|---|---|
| The Signature Was Valid. The Authority Wasn't. | Who said “go” is not the same question as who was allowed to. |
| The Human Approved It. Which “It”? | Does the approval still bind to the action that actually executes? |
| The Tool Was Read-Only. The Consequence Wasn't. | Is the safety property enforced—or merely attached to a name? |
04 / CLEAR MENTAL MODELS
If the explanation cannot survive a curious kid, keep working on it.
Robot Explains turns web, cybersecurity, and AI concepts into illustrated explanations, guided learning, and games. What is it? Why does it matter? How do I use it safely?
Side channel — sometimes the explanation is satire
Bigger Sandbox — Containment, but bigger this time.
Privacy™ — Search anyone. Know everything. Be wrong at scale.
Satirical sites about containment theater and surveillance overconfidence. The joke is the interface. The subject is the assumption underneath it.
05 / FORMAL QUESTIONS
Keep separate questions separate. Make the relationships explicit.
MICRM is a formal reference model for machine-initiated consequence attribution. It distinguishes consequence structure, authority and accountability closure, contribution-sensitive candidacy, and evidentiary warrant rather than silently substituting one for another.
Scope matters: this is a reference model with stated limits—not a completed theory or a universally validated, substrate-neutral framework.
Research index — adjacent models and control questions
Frontier Cyber Risk Taxonomy — cyber-assistance classification across uplift, autonomy, authorization verifiability, and cumulative capability transfer.
Agent Security Threat Model — trust boundaries across agents, tools, browsers, files, APIs, and cloud resources.
Agent Tool Permission Matrix — default-deny grants tied to risks, required controls, and enforcement points.
Safety-Router Transparency — useful explanations without turning disclosure into an oracle for an attacker.
MAIN SITE / FIELD NOTES / ROBOT EXPLAINS / LINKEDIN
Independent public work. No employer endorsement implied. Claims travel with their scope, limitations, and evidence.