Writing

Securing AI-DLC: Process Integrity Is Not Product Security

AI-DLC gives you the strongest process integrity I have seen in any methodology — approval gates, audit shards, element-level traceability. None of it tells you the code is safe. Those are different claims, and closing the gap phase by phase is the work.

Monday, August 31, 2026 | 10 minutes Read

The Security Coach for AI Agents

Every team debating whether to trust the model has already made the decision that matters: where the boundary lives. A prompt is a promise. A tool interface is a constraint. Four patterns for telling them apart.

Tuesday, August 25, 2026 | 8 minutes Read

Sitadel: What You Owe People Once a Distro Ships Your Tool

Eight years, 179 commits, and packaged in Kali and BlackArch — which I did not plan and which changed what I am allowed to break. On finding quality over detection coverage, a hang that only existed in a real terminal, and the cleanup nobody asks for.

Tuesday, August 18, 2026 | 7 minutes Read

Watari: A Security Control That Was Never On

Watari’s headline claim is tenant isolation enforced by PostgreSQL Row-Level Security. The policies had been in the schema for months and had never once taken effect. Turning them on took four changes that had to land together — and none of them was writing a policy.

Tuesday, August 11, 2026 | 9 minutes Read

Secrust: When Adding Cores Made It Slower

A single-process Sigma correlation engine in Rust. Eight optimisation passes took the realistic workload up 301% — and then four threads made it 23% slower than two. Why that happened, what I shipped as the default, and the things I decided not to build.

Tuesday, August 4, 2026 | 9 minutes Read

Cutting 90% of the Noise: Agent-Driven Security Operations

Tuning detections until the volume fits your headcount means choosing your visibility based on your staffing. The fix is not more analysts — it is separating the three layers of triage by what each one is allowed to be wrong about.

Wednesday, July 8, 2026 | 7 minutes Read