Hypervisor
A single physical machine runs about ten isolated systems. A backup is taken before every significant change.
Two separate setups. On one side, the server that hosts this site. On the other, a private lab split into sealed zones, where I practice attack and defense safely.
no communication between zones
This site runs on a server reserved for it, with no link to the lab. Each part has a single role.
Linux server hosted in France. Restricted administrator access, automatic security updates.
A firewall denies everything by default and only opens what is strictly needed. Repeated login attempts are blocked automatically.
Traffic is encrypted end to end, with a certificate renewed automatically. The browser receives strict rules on what it is allowed to load.
A single command updates the site, from the repository to production. Every version is rebuilt identically, with no manual step.
The libraries used by the site are watched automatically. When a flaw is published, the fix is proposed the same day.
The site never receives requests from the Internet directly, they all go through the proxy. It runs with minimal rights and capped memory.
A second-hand server brought back into service, with limited memory. That constraint shaped the whole architecture, and it is what makes the exercise interesting.
A single physical machine runs about ten isolated systems. A backup is taken before every significant change.
A firewall splits the network into four zones that do not talk to each other. None of them can be reached from the Internet.
A detection system watches the traffic. I attack from one zone and check that the alert really fires in the other.
A rebuilt company directory, like the ones found in most organizations. It is used to understand how these environments are attacked and defended.
A machine dedicated to penetration testing, facing deliberately vulnerable applications. It can reach neither the rest of the lab nor the Internet.
Administration goes through an encrypted tunnel, with a different key for each device.
The lab watches the site from the outside. A daily check compares its state with the day before and reports the slightest difference.
Does the site answer, is the connection properly encrypted, is the certificate still valid.
Are the protections sent to the browser still in place, and has the page content changed for no reason.
Does the domain name still point to the right place, and has nobody else obtained a certificate in my name.
The next step is to rebuild this architecture at home, on my own network. The hardware is being gathered.
A salvaged machine, brought back into service to host a second lab, independent from the first.
A custom-built storage server, to keep backups somewhere other than the machine they protect.
A firewall at the edge of the home network, to keep personal use apart from the lab.
What follows is not a roadmap, it is a direction. I want to design everything I build, from the circuit board to the software that drives it, all the way to the finished object. Then connect these fields in a single infrastructure, with shared tooling able to cover them all, and that stands on its own without owing anything to anyone.
Some of these subjects are out of reach today. They will stay that way for less time if I prepare for them now.
Devices that prove what they run, instead of asking to be believed.
All security today rests on trust granted to the manufacturer. I want to reverse that. Design hardware that can show the person using it which code it really runs. Doing it in software is already hard. Doing it in silicon is the next step.
Protecting devices whose failure cannot be undone.
A monitor, a pump, an implant. These machines keep people alive and are sometimes controlled without serious authentication. Regulation is moving, tooling is not. Everything is still to be built, and almost nobody is working on it.
Equipment where every component is under control, from the circuit to the code.
A weapon system that cannot be fully audited remains a disguised dependency. Designing, building and verifying on one's own territory is not a stance, it is the only way to know what the things you deploy really do.
Securing the machines that keep a country running.
Water treatment, energy, production lines. Thirty-year-old controllers that can be neither stopped nor replaced, now connected to networks they were never designed to face. Protecting them without touching them is a beautiful problem.
Machines that move and decide on their own, and can therefore be hijacked.
Drones, robots, vehicles. An intrusion there no longer steals data, it takes control of an object that acts in the real world. Knowing how to design them and knowing how to defeat them are two sides of the same craft, and I want both.
Seeing, understanding and contesting what travels over the air.
Radar, telemetry, command links. Anything that transmits can be detected, analyzed and disrupted. This invisible battle takes as much electronics as signal processing, and it is one of the most demanding fields of the discipline.
Building the missing devices needed to observe hardware.
Auditing a board means reading its signals, its power draw and its memory while it runs. The instruments that do this cost tens of thousands of euros, when they exist at all. Building them yourself opens doors that nobody else opens.
Attacking and defending systems that learn.
A model embedded in a machine becomes a new kind of attack surface. It can be fooled, extracted, poisoned. In return, it spots weak signals that no hand-written rule would catch. Both directions interest me equally.
All of it with my own hands.