Ethernet cables plugged into a network switch

← All services

Breach prevention and hardening

Most breaches use something boring: an exposed admin page, a forgotten service, a server that can call anywhere. I close those doors, one logged change at a time.

Hardening: what it covers

  • Servers, containers and web servers: configuration, exposed services and admin paths.
  • Network and firewall rules: what may talk to what, and what is simply denied.
  • Egress control: allow-listing outbound traffic so a compromised host cannot phone home.
  • Access: least privilege, individual credentials, strong authentication on administrative paths.
  • Secrets, backups and restore: that they exist, are protected and actually restore.
  • Patch posture: what is out of date, and in what order to fix it.

What a check looks like

hardening check · example hostsample values
PASSPHP execution denied in upload directories
PASSLegacy remote-call endpoint returns 403
PASSOutbound traffic limited to an allow-list
PASSAdmin paths behind a one-time-passcode gate
FAILDatabase listens on all interfaces
UNKNOWNBackup restore: could not verify

Checks only read. Fixes come as a script plus a rollback, never applied automatically.

How it runs

  1. Assess

    Read-only review of configuration and exposure. Nothing changes.

  2. Plan

    A ranked list of changes, each with its purpose, its risk and its rollback.

  3. Apply

    In agreed windows, after your approval, staged from least to most critical systems. Hardening is generated and checked, never applied blindly: a person reads the change before it runs.

  4. Verify and record

    Each change is tested from both sides (what should work, what should now fail) and entered in a register with where it lives and how to undo it.

Testing: what it covers

I have run penetration tests for multiple clients over the years, including Obsentum and other companies in its group.

  • Attack-surface review: what is reachable from the internet, including things you forgot you ran.
  • Web applications and APIs: authentication, authorisation, input handling, session and data exposure.
  • Infrastructure and network: exposed services, weak configuration, trust between systems.
  • Identity and access paths: how an attacker would move from a first foothold to something valuable.
  • Retest: confirming that fixes work.

Ground rules

  • Testing starts only after a signed scope: targets, methods, windows, contacts and stop conditions.
  • No destructive testing, denial of service or data extraction beyond proof, unless you agree it explicitly in the scope.
  • Anything critical is reported to you immediately, not saved for the report.
  • Test accounts and access are individual, time-limited and removed at the end.

Monitoring that earns attention

  • State-change detection: new listening ports, new accounts, changed system files and modified web roots.
  • Pages only for changes that favour an attacker; everything else goes to a periodic digest.
  • Detect-only agents for appliances such as firewalls: they look and report, they never change a rule.
  • A tamper-evident, hash-chained evidence log, so records cannot be quietly edited afterwards.

Staged patch rollout

Systems are grouped by sensitivity: Lab, Normal, Important, Critical. A new version reaches a group only after every lower group has run it without failure, with a pause in between. Test and internal machines always go first.

  1. Lab

    Test and internal machines take the change first.

  2. Normal

    Staging and pre-production follow once Lab is clean.

  3. Important

    Live production, after a pause.

  4. Critical

    Payments, billing and the main database, last, once everything below has proven it.

  • Updates are signed, and checked for size, hash and signature before anything is touched.
  • If the service fails after an update, the previous version is restored automatically.
  • A platform or host can be paused or pinned when a release should wait.

What it looks like

staged rollout · agent updatesample values
Labdone
Normaldonepause 30 min before the next level
Importantrunning
Criticalqueued

On failure the previous version is restored automatically. Hosts can be paused or pinned.

evidence log · hash-chainedsample values
#0411a41f0c92 ← 7be3d508 host check-in
#04125d27e8b1 ← a41f0c92 new listening port: alert
#0413e90c3a76 ← 5d27e8b1 update verified, applied

Edit any row and every later link stops matching. Verification fails.

Built on a tool I use myself

The approach comes from fleet-watch, the tool I built to watch and patch servers: break-in detection, signed update jobs, staged rollout by sensitivity level, automatic rollback and a hash-chained evidence log. I run it on the servers I am responsible for, clients' included, and offer it as a managed service. See fleet-watch.

What you receive

  • A prioritised findings list in plain language.
  • A hardening register: every change, its location and its undo path.
  • Verification evidence you can hand to an auditor.
  • A short debrief and a list of what to watch next.

A good fit when

You run internet-facing systems or sensitive internal ones, you suspect the basics are not all in place, or a customer or regulator has started asking how you protect them.

Want to know which doors are open?

I will tell you what I would look at first.

Get in touch