Back

Mechanics

Complete explanations of how things work — from first principles to full understanding.

Series
Articles

The Authorization Problem: Why LLMs Cannot Guard Authentication Boundaries

When an AI agent is granted write-level access to account management functions, the security model collapses — not because the model is poorly designed, but because language models are architecturally incompatible with the trust assumptions authentication systems depend on.

How ptrace Works

ptrace is the only mechanism Linux provides for one process to observe, interrupt, and modify the execution of another. Every debugger, syscall tracer, and container runtime syscall filter uses it. Here is how the kernel implements it and where the access model breaks down.

How Windows Privilege Escalation Works

Windows enforces privilege boundaries through integrity levels, access tokens, and the Security Reference Monitor. Privilege escalation exploits don't break that model — they find paths through it. Here's how the enforcement works and where the seams are.

How OIDC Trusted Publishing Works — and Where Mini Shai-Hulud Found the Gap

OIDC trusted publishing was designed to eliminate the long-lived credentials that supply chain attackers steal. Mini Shai-Hulud bypassed it anyway. Here's how the mechanism works, what it actually guarantees, and how three individually reasonable configuration decisions combined to let an attacker publish under TanStack's own verified identity.

Why a Linux Binary Won't Run on Windows

A program compiled for Linux won't run on Windows, and a macOS binary won't run on Linux. The reason isn't the code — it's what the binary expects the operating system to do for it.