Luminana

AI security experiments
& red-team notes

MEM 0x00194000 VER 2.0 / EST. 2025 FOCUS: LLM / KERNEL / 0-DAY

[ SPECIMEN 01 — DRAG TO ROTATE / CLICK TO MUTATE ]

A working notebook for breaking intelligent systems — documented as field notes, not marketing.

FIG. 02Abstract

// hypothesis

Every guardrail is a claim. Luminana exists to test the claim — with adversarial prompts, bent kernels and patient curiosity, one experiment at a time.

Findings are written up on the blog as reproducible notes: what was tried, what broke, and what it implies for people building — or defending — these systems.

FIG. 03Research Domains

01 AI Model Security guardrails / jailbreaks → 02 Prompt Injection adversarial inputs → 03 Kernel & Systems low-level tinkering → 04 Zero-Day Research undisclosed vulns →

FIG. 04Research Index

001

FIG. 05Operators

OP-01

a0x194

— AI model pwn — prompt inject

OP-02

Roy

— kernel panic — zero-day

Prompt Injection: Mapping the Attack Surface

When Guardrails Bend: Notes from Model Testing

A Kernel Panic Postmortem

Field Notes: Building a Home Red-Team Lab

a0x194

Roy