The quoting app has been running without trouble for two years. Then one day a salesperson notices that if he changes the number at the end of the URL, he can see another rep's quotes. Including the competitor quotes someone uploaded for reference. Nobody had tried it because nobody thought they needed to.
Web application security in custom software rarely breaks because of a sophisticated attack. It breaks because of things like that: a control everyone assumed was there and nobody wrote into the requirements. When you commission a build, the vendor usually quotes screens and workflows. What you can't see (who can sign in, what each person can do, what gets logged) depends on you asking for it, or on the team bringing it out of habit.
What follows is a list of minimums: what any web application built for your company should ship with from version one, using the OWASP Top 10 as the reference. A full cybersecurity program is a different conversation. If the system was written with AI coding assistants, there are extra risks we cover in AI software development: code risks.
The OWASP Top 10 as a starting point
OWASP is a nonprofit foundation that periodically publishes the most common risks in web applications. The current edition is the OWASP Top 10:2025, which replaced the 2021 list. It works as a shared vocabulary: ask a vendor how they handle each item and their answers will tell you a lot about how they work.
| OWASP 2025 risk | In plain terms | Minimum control |
|---|---|---|
| A01 Broken Access Control | A user sees or changes things they shouldn't | Check permissions on the server, on every request |
| A02 Security Misconfiguration | Default settings, exposed admin panels, error messages full of technical detail | Reviewed configuration per environment, no test accounts in production |
| A03 Software Supply Chain Failures | A vulnerable or tampered third party library | Dependency inventory and regular updates |
| A04 Cryptographic Failures | Sensitive data unencrypted or using weak algorithms | HTTPS everywhere, properly hashed passwords, secrets out of the code |
| A05 Injection | User input that ends up executed as code or a query | Parameterized queries and input validation |
| A06 Insecure Design | The flaw is in the logic, not in a line of code | Think through abuse cases at design time, not just use cases |
| A07 Authentication Failures | Weak passwords, sessions that never expire, no second factor | SSO or MFA, lockout after repeated attempts |
| A08 Software or Data Integrity Failures | Updates or data accepted without checking where they came from | Controlled, signed deployment pipeline |
| A09 Security Logging and Alerting Failures | Something happened and there's no way to know what or who | Audit log and alerts on critical events |
| A10 Mishandling of Exceptional Conditions | Errors that leave the system in an unsafe state | On error, deny access instead of allowing it |
Authentication: SSO and MFA from day one
The first minimum is not to invent your own user system when the company already has one.
If your organization runs Microsoft 365, it already has Microsoft Entra ID as its directory. The application can hand sign-in over to it (SSO) and inherit the multifactor authentication and policies IT has already set up. When someone leaves and their account is disabled, they also lose access to the custom app, and nobody has to remember to do it.
When the app has external users (customers, suppliers, contractors) and the corporate directory isn't an option, these are the minimums:
- Passwords stored with a hashing algorithm designed for it (bcrypt, scrypt or Argon2), never in plain text or with a generic hash.
- A second factor available, and mandatory for privileged accounts.
- Lockout or progressive delay after several failed attempts.
- Sessions that expire and are invalidated on sign-out or password change.
Authorization: defined roles, checked on the server
The opening scene is a broken access control case, the top risk on the OWASP list. It happens when the app hides a button on screen but the server never checks whether the person making the request is allowed to make it.
At minimum, the requirements should include a role matrix: who can view, create, approve and delete each type of record. With that matrix the development team builds the checks into the backend, on every operation, and QA has something to test against. During acceptance testing, ask someone with a low-privilege role to try what they shouldn't, for example by changing IDs in the URL. If it works, the app isn't ready.
Inputs, secrets and dependencies
Validating input means treating everything that comes from the user as suspect: forms, uploaded files, URL parameters and integrations with other systems. Database queries should be parameterized, and uploaded files should be checked for type and size before they're stored.
Secrets (API keys, database passwords, certificates) don't belong in the code or the repository. They go in a secrets store, such as Azure Key Vault if the app runs on Azure, or in environment variables managed by the operations team. If a key ever ended up in Git history, rotate it. Deleting the file isn't enough.
Dependencies are the code your application uses that nobody on your team wrote. OWASP moved them up to third place in 2025 with the software supply chain failures category, and its recommendations include keeping a component inventory (SBOM), tracking indirect dependencies too and removing the ones you don't use. In practice, ask for automated dependency scanning on the repository and for someone to actually read its alerts.
Audit logs and backups
When something goes wrong, the first question is what happened, and without logs the answer is "we don't know." A business application should record, at a minimum, sign-ins (successful and failed), permission changes, creation, edits and deletion of important records, and bulk data exports. Each event with user, date, time and source.
Those logs should live somewhere the app's own users can't delete them, and it's worth setting alerts for things that should never happen, like one user downloading thousands of records in an afternoon.
Backups have the same problem: having an automatic one means nothing if nobody has ever tried restoring it. Ask for the contract or operations plan to state how often backups run, where the copy is kept and when the last restore test happened.
There's a legal angle for Chilean companies too. Law 21,719 requires anyone responsible for personal data to adopt security measures appropriate to the risk. It's set to take effect on December 1, 2026, and a bill in Congress would push that to October 15, 2027, which, according to Eguiguren Abogados' tracker as of November 6, isn't final yet. Delay or not, audit logs and tested backups are part of how you show you did what was reasonable.
How to ask for it and how to check
All of this is easier to request at the start than to add later, when it can mean rewriting parts of the application. Before you sign, put these in the requirements:
- Authentication method (SSO with the company directory or, if that doesn't apply, built-in MFA).
- Role and permission matrix.
- Policy for handling secrets and dependencies.
- Events that must be logged and who can view them.
- Backup frequency and a restore test before go-live.
- A security review before launch, using the OWASP Top 10 as the checklist.
If the system already exists and you don't know what state it's in, the way forward is an outside review. We explain how in software development consulting: code audit.
Frequently asked questions
Is the OWASP Top 10 a certification?
No. It's a reference document on the most common web application risks. No app gets "OWASP certified." What you can ask is for the vendor to show how they handle each category.
Do I need MFA if only employees use the app?
Yes. If the company already uses Microsoft Entra ID, the simplest route is to connect the app through SSO and apply the same MFA policies that already cover email and files.
How often should a custom application's security be reviewed?
Before going to production and after major changes, such as new modules, integrations or external users. In between, automated dependency scanning and audit alerts handle the day to day.
Next step
At Disrupsoft we build and integrate applications with secure APIs, SSO and MFA, and connect existing systems to other platforms. If you want to review the security minimums of an app you're about to commission, or one you already have, let's talk at Integration and Security.



