Skip to content
M&L AI

About

About M&L AI — AI-driven defensive software development, plus offensive security testing and red-teaming.

M&L AI logo

M&L AI is where I make AI-driven software with a focus on defensive cybersecurity. In addition to defensive, AI-driven software development, I also do offensive security work: offensive software testing and development, vulnerability testing, and red-teaming. I’m a privacy advocate and software developer, and everything I build here is open source.

My work centers on a simple idea: software should respect the people who use it. That means privacy by default, no surveillance, no dark patterns, and code you can read, audit, and run yourself. I care about giving people practical control over their data and their tools.

How I work

The way I build — small, protocol-first, status-honest, documentation-close-to-code — isn’t an accident. It comes from a few convictions:

Privacy is a default, not a feature. The tools I build don’t phone home, don’t track you, and don’t collect what they don’t need. Local-first wherever it’s possible.

Security architecture is design work, not paperwork. Threat models and protocol RFCs are first-class artifacts, written before the code stabilizes, not retroactively to satisfy a checklist.

Status you can trust beats marketing. Everything I ship carries an explicit status. “Alpha” means alpha. “Hardening pending” means I haven’t yet hardened it. I’d rather lose a lead than oversell a v0.1.

Composability over monoliths. I favor open protocols and module boundaries that let other people swap pieces, and standard formats over bespoke servers. Costs more upfront and pays off everywhere downstream.

See build principles for the longer version.

What I build

AttackMap is a local-first, open-source defensive security analysis engine. It reads a repository’s source code, reconstructs the attack surface — routes, data stores, trust crossings, secrets in the wrong places — and produces prioritized, evidence-anchored findings mapped to MITRE ATT&CK. Code never leaves the machine, and the output is oriented entirely toward remediation: no exploit code, no offensive payloads. Full documentation lives at docs.mlaify.io.

Looking for help AttackMap is looking for contributors and co-maintainers. It’s under active development, but progress may be slow until more help or co-maintainers join — if you’d like to help with the core engine, an analyzer, the macOS app, or the docs, open an issue to say hello.

grrclone is a free, open-source macOS menu bar app that connects rclone remotes as Finder volumes. It auto-discovers remotes, reconnects at login, and repairs mounts after sleep or a network change. No licence key, no phone-home, no kernel extension, no root. Install it with brew install --cask mlaify/tap/grrclone.

openlens-fw is hardened, open-source replacement firmware for the Anpviz IPC-D240W-S IP camera, written after the camera security research found 24 vulnerabilities in its stock firmware. It keeps the stock kernel and replaces the whole userland: no shared credentials, nothing that parses the network running as root, a certificate authority on each camera, and updates signed with the owner’s own key. Its documentation is at docs.mlaify.io.

Offensive security

Defensive tools are only as good as the attacks they’ve been tested against, so I also work the other side: offensive software testing and development, vulnerability testing, and red-teaming. I take on offensive work only in authorized engagements. The Anpviz camera security research is one example: testing the camera’s stock firmware found 24 vulnerabilities, which I reported through coordinated disclosure, and the hardened replacement firmware above grew out of it.

Elsewhere

  • matthewd.xyz — my personal site and writing
  • GitHub — code
  • Docs — documentation for AttackMap and the other projects