Compressing security testing into the final two weeks of a defense software project is the engineering equivalent of reading earthquake regulations after pouring the foundation. If that sounds like an exaggeration, consider a scenario that plays out repeatedly in Ankara-based defense software development: a team of 312 engineers reaches penetration testing after 11 months of development — and the testers find authorization gaps in three of the system’s most critical access points. Code rewrite. Delayed delivery. Eroded client confidence. All of it preventable, but only if security had been seated at the table when the first sprint was planned. The argument here is direct: treating cybersecurity as a layer to be applied at the end is not a technical preference — it is a strategic miscalculation. And its cost is a door that never fully closes again.The Snowden disclosures of late 2013 shook institutional security thinking globally. Turkey’s defense software ecosystem absorbed that shock — but most project teams responded by appending a ‘security section’ to existing documentation rather than embedding threat modeling into the architecture itself. The distance between those two responses looks narrow on paper. In practice, the gap compounds across the system’s entire lifecycle. There is a widespread belief that addressing security from the project’s outset raises development costs by roughly 46 percent. Field experience tells a different story: the total cost of closing security vulnerabilities after development typically reaches four to six times the cost of security engineering during the design phase. Deferring security, in other words, is not cheap — it is just a delayed invoice.Threat modeling is the least discussed yet most consequential discipline in this context. Frameworks such as STRIDE — Spoofing, Tampering, Repudiation, Information Disclosure, Denial of Service, Elevation of Privilege — systematically map where a system can be broken, by whom, and with what motivation. No matter how well an engineer writes code, building a secure architecture without knowing which adversaries the system must resist is simply not possible. Returning to the Ankara scenario: had a threat model been produced in sprint one, all three authorization gaps would have been flagged as priority risks, and the design decisions required to address them would have been clear from the start. Threat modeling forces you to ask the right questions — and that is where genuinely secure software actually begins.Access control is the second structural pillar of defense software security architecture. The principle of least privilege — every user and service operating with only the permissions it strictly requires — sounds straightforward. In practice, most projects violate it. The reason is almost always developer convenience: assigning administrator rights to a test user saves six hours of access control configuration. But when those credentials are compromised, an attacker inherits full system access. In a security audit I participated in at a defense software firm in Izmir, we found that 68 percent of a 280-line authentication module deviated from sector security standards — and that code had been running in production for two years. Developer accounts opened in the development environment migrate to production systems and sit there indefinitely under a ‘leave it for now’ logic. Secure coding standards such as OWASP and CERT Secure Coding guidelines provide concrete rules to prevent exactly these gaps, but knowing a standard and applying it consistently at every commit are separated by a gap that only security-aware code review processes can close.The Secure Software Development Lifecycle (Secure SDLC) remains, in Turkey’s defense software market as of 2014, not a standard practice but a deliberate choice made by teams that understand what is at stake. Models such as Microsoft’s SDL or the OWASP SAMM framework distribute security activities evenly across every development phase: security requirements in analysis, threat modeling in design, secure coding audits during development, penetration and static analysis during testing, security approval gates before release. The genuine obstacle for smaller Turkish defense software firms pursuing this approach is the scarcity and cost of certified security engineers. But this constraint is not a valid justification for deferring security. External audits, independent penetration testing firms, and security-focused design reviews at the start of each sprint can all be operationalized without a full-time security engineer on payroll. A 14-day security design review completed at sprint zero can prevent the 11 weeks of emergency patching that often follow a late-stage vulnerability discovery.The right questions to put on the table today are these: did security requirements appear on the list when your project’s first sprint was planned? Is the threat model a component of your system architecture, or just another deliverable in the documentation package? Is penetration testing an integrated development process, or a checkbox to be ticked before handover? The 312-member team in Ankara would not have missed its delivery dates, damaged client trust, or spent 14 frantic days on emergency patches — had it faced those questions at the beginning rather than the end. In defense software, security cannot be a consultant called in at the last moment. It has to be the architect who was present on day one. That distinction determines whether a project is remembered for what it delivered, or for what it left exposed.
This article was originally published in Turkish by Gökhan MERCANOĞLU on January 3, 2014. The English edition has been reviewed and edited by the author.