Custom Software2026-11-17

Custom Software Security: 10 Essentials for Every Project

Security in custom software is cheaper when designed in than when retrofitted. This guide covers the ten essentials every project needs, from threat modelling and dependency policy to secrets handling, audit logging, backup restoration tests and release gates.

Custom Software

Why custom software is a different security problem

Off-the-shelf software inherits a vendor's security programme: patching cadence, penetration tests, disclosure process. Custom software inherits nothing by default, so every control must be a deliberate decision made by your team. That is the core risk of building your own: the responsibility for patching, monitoring and access design moves to you. The ten essentials below are the minimum set that turns an ordinary custom build into something defensible in a security review, and they are grouped into four families: design-time controls, data and dependency hygiene, operational resilience, and release discipline.

1-3. Threat modelling, identity design and authorisation

Threat modelling answers three questions per feature: what are we protecting, who would attack it and how. A one-hour session per major workflow, using a simple list of assets, entry points and trust boundaries, catches issues that are cheap at design time and expensive later. Use a proven identity provider or library rather than hand-rolled authentication; enforce multi-factor for administrative roles, rotate session tokens on privilege change and set absolute as well as idle timeouts. Authentication proves who someone is, but authorisation decides what they may do and must be enforced server-side on every request. The most common custom-software breach is object-level authorisation: an API returning a record because the identifier was guessed. Adopt a pattern where each endpoint declares the permission it requires and middleware enforces it, then write automated tests that attempt cross-account access.

4-6. Hostile input, secrets and encryption

Treat every input as hostile, including headers, file names and webhook payloads. Validate against an allow-list, reject rather than sanitise when the value is structurally wrong, encode output for its context - HTML, attribute, URL, JSON or SQL - and use parameterised queries exclusively. For uploads, inspect the file rather than trusting the extension, store outside the web root and serve through a handler that sets the correct content type. Secrets belong in a managed store or environment injection, never in a repository, build log or screenshot; rotate on schedule and immediately after any suspected leak, with a pre-commit hook and repository scan to catch accidental commits. Pin dependency versions, enable vulnerability alerts, maintain a software bill of materials so a critical advisory can be answered within an hour, and classify data before encrypting - in transit with modern protocols, at rest for anything sensitive, with keys managed separately and retention automated.

7-8. Audit logging and restore testing

Log security-relevant events - authentication, privilege changes, administrative actions, exports, deletions - with actor, timestamp, source and outcome, ship them to storage the application cannot modify, and set retention to match legal obligations. Build alerts for patterns rather than single events: repeated failed logins from one address, a spike in export volume, an administrative action outside business hours. Test the alerting path quarterly with a benign event, because an alert nobody receives is worse than none. A backup that has never been restored is a hypothesis, not a control: define recovery objectives in numbers, test restoration quarterly into an isolated environment, record the actual time taken, keep one copy offline or immutable against ransomware, and document the recovery sequence including who declares an incident and who talks to customers.

9-10. Release gates and practised incident response

Separate development, staging and production with distinct credentials and no production data in lower environments unless masked. Automate builds so every release is reproducible and traceable to a commit, require peer review for changes touching authentication, payments or data access, and run dependency and secret scans as a release gate. Define a rollback procedure someone has rehearsed, and keep an emergency path for critical patches that bypasses scheduling without bypassing review. Alongside that, write a one-page incident plan: who is called, in what order, with what authority, and what is communicated to customers and regulators within which timeframe. Keep contacts offline as well as online, practise a tabletop exercise twice a year, and write a short post-mortem after every real or simulated incident naming the root cause and the specific control that failed.

How to sequence the work across two quarters

Attempting all ten essentials at once usually produces a plan nobody executes. Sequence by risk instead. Quarter one: threat model the two workflows that touch money or personal data, enforce multi-factor on administrative accounts, add server-side authorisation tests, and run a secrets scan in the pipeline. Quarter two: ship audit logging with a tested alert path, run the first real restore drill, define the release gate, and hold one tabletop exercise. Everything else - encryption key rotation, dependency automation, accessibility of admin interfaces - follows once these are verifiably in place. Each step should end with evidence you can show a reviewer: a document, a passing test, a logged drill with a timestamp.

A minimum baseline you can audit

If you implement nothing else this quarter, implement five things: an inventory of every service that stores data and who can access it; multi-factor authentication on all administrative accounts; server-side authorisation tests in the automated suite; a secrets scan in continuous integration; and one tested restoration from backup. These five cover the failure modes that appear in almost every breach report, and each can be completed in weeks rather than quarters. Measure them as yes-or-no facts in a review, not as intentions, because a control that has not been executed and verified does not exist.

Get a free security design review

Building custom software and want an independent read on its security design? Book a free review with our team. We will examine your threat model, authentication and authorisation design, dependency policy and logging plan, then give you a prioritised list of gaps with effort estimates. You leave with a plan you can execute before launch.

Need Help With Your Project?

Get a free quote and get a fixed quote within 48 hours.

Get a Free Quote