Skip to content
Tech Interview Prep
Technical interview guide

Secure Software Development Lifecycle

Building security checks into every phase of development instead of testing for it only at the end.

Read
30 min
Practice MCQs
25
Interview Q&A
25
Edition
v4
Editorial status
Reviewed

Scope: NIST SSDF 1.1; SLSA v1.2; OWASP SAMM and ASVS current 2026-08-31; CISA Secure by Design and Product Security Bad Practices current 2026-08-31.

Curated: · Written: · Reviewed:

25 questions

MCQ-3A DoD software attestation requires each SDLC activity to be mapped to a top-level practice group of NIST SP 800-218 (SSDF v1.1), with the headings matching the publication. The draft sheet uses headings borrowed from other frameworks. Which four headings belong in the SSDF column?
MCQ-7A payments team is adding a required check to its pull-request pipeline. The check must run on every commit, cannot reach a deployed environment or attach an agent to a running JVM, and must report a dataflow from an HTTP parameter to a SQL query in the Java source. Which analysis does that check perform?
MCQ-19A CI pipeline builds container images, its job output is readable by every engineer, and images are published to a shared registry. Which practice keeps credentials out of source control, build output, and published artifacts?
MCQ-11Release 4.2.0 ships with a CycloneDX SBOM generated at build time from the resolved dependency graph: 412 components, each with a package URL and version, including log4j-core 2.14.1. The SBOM carries no runtime classpath, call-graph, or configuration data. A CVE is published for log4j-core before 2.15.0. Which conclusion does the SBOM alone support about release 4.2.0?
MCQ-12After a compromised dependency shipped in an image that passed their SLSA v1.0 provenance gate, a platform team reviews what the gate actually checked: the attestation signature against the org's build-service key, `runDetails.builder.id`, and the `subject` digest against the image. Which of these does the verified provenance not establish?
MCQ-2A vendor's appliance ships with the same factory-default admin credential on all 12,000 deployed units. Telemetry shows 38% still authenticate with that default 90 days after install, and 70% of units are provisioned through the CLI/API rather than the setup wizard. Under secure-by-design principles, which change should the vendor make?
MCQ-1A team reviewed 50 production security defects from the past 18 months and classified each by the earliest phase that could have caught it: 31 in design (missing or incorrect security requirements), 12 in implementation (a known-unsafe API used as documented), 5 in a third-party dependency, and 2 in deployment configuration. They already run SAST on every pull request, which flags the 12 implementation defects, and a pen test before each release. Which single change removes the largest share of these defects?
MCQ-4A team is turning the security requirement 'refund requests must be authorized' into an acceptance test. Their harness can mint actor tokens with any scope set, call the refund endpoint, and read only the HTTP status code and JSON response body. It cannot create users, assign roles, read logs or the database, or call the policy service directly. Which rewrite of the requirement can this harness verify as written?
MCQ-10A Go service builds a user lookup by concatenating the search box text into a statement: SELECT id, email FROM users WHERE name = '<input>'. A hardening pass replaces every ' in the input with '' before concatenation and leaves backslashes alone. The database is MySQL 8.0 on its default sql_mode, where backslash escapes the next character inside a string literal. Which input does the quote doubling fail to neutralize, letting the query return every user's row?
MCQ-8A Node.js 20 / Express 4 service exposes `GET /api/invoices/:id`. The handler runs `requireAuth`, which verifies the JWT signature and expiry and nothing else, then calls `db.invoices.findById(req.params.id)`, a parameterized query. SAST scans the route in CI and reports no high-severity findings. The nightly DAST scan runs authenticated as user A, reaches the endpoint, and requests the invoice ids linked from A's own pages; it reports nothing. A pentest against the same deployment then shows user A can read user B's invoice by requesting B's id. Which explanation accounts for both tools missing it?
MCQ-16An API gateway caches authorization decisions for 30 seconds in a dict keyed by `(user_id, request.path)`. The decision function is `can_access(user_id, method, path)`. A PR adds the cache and a test asserting that two `GET /orders/42` requests from the same user get the same decision; the test passes. What must the review block on?
MCQ-17A multi-tenant SaaS API takes an object id and a tenant id on every endpoint, and orders move through draft → submitted → approved. The last release shipped a bug where a viewer in tenant A approved an order belonging to tenant B. SAST, dependency scanning and a TLS configuration review were clean. Which test would have caught it?
MCQ-5A threat-modeling session for a payments webhook receiver records one path: an attacker who captures a signed `payment.succeeded` body can POST it again and the receiver credits the account a second time. The receiver checks a processed-event table before crediting, but it inserts the event ID in a second transaction that commits after the credit transaction. The team has one sprint before launch and can ship one change. Which change closes the modeled path?
MCQ-6A team ships an on-prem appliance. Of 4,000 field installs, 1,080 have the admin console reachable from the internet: the installer binds it to 0.0.0.0 and the quickstart's first step is "open https://<host>:8443". Support data shows 90% of those 1,080 installs never edited a config file after install. The team has one release before an audit. Which change cuts the 1,080 figure the most?
MCQ-13A CI system runs 400 PR builds a day on 12 long-lived runners. Each runner mounts a service-account key with write access to the production artifact registry, and the key stays valid for 90 days. A contributor's PR edits the build script; the next build overwrites a production artifact with a tampered copy that is served for six hours. The team wants the smallest change that would have stopped this build from writing to the registry. Which change?
MCQ-21`pathkit` is an internal library that each service copies into its own source tree rather than importing from a registry. A path-traversal defect is present in `pathkit` v2.0.0 through v2.4.0; v2.4.1 fixes it. The vendored versions in production today are: checkout 2.4.0, billing 2.1.0, search 2.4.0, reports 2.0.0, admin 1.9.0, auth 2.4.1. The team re-vendors 2.4.1 into checkout, redeploys it, and closes the incident. Which services still ship the vulnerable code?
MCQ-14A Node service ships as a container built with `npm ci --omit=dev`. The lockfile resolves 640 packages: 11 direct and 629 transitive. A CVSS 9.8 CVE is published in `png-turbo`, which the lockfile shows is required only by `jest-image-snapshot`, a devDependency. The next release is in 6 hours. What should the team do before the release?
MCQ-15A scanner reports SQL injection in the /search endpoint of a public API. The team files a suppression with a compensating control: 'the WAF blocks SQLi.' The only WAF rule covering /search matches the literal string 'UNION SELECT' in the query string. The exploit uses a time-based blind payload: `1' AND SLEEP(5)--`. The scanner re-runs nightly. What is the defensible outcome?
MCQ-18A payments team's release gate says: any critical or high finding in code reachable from an internet-facing entry point blocks the release unless (a) the reachable path is disabled in the shipped configuration, (b) a compensating control is deployed, and (c) an exception with a named owner and an expiry no later than 30 days is recorded. Medium findings need a named owner and an expiry but do not block; low findings are logged only. Release candidate RC-14 has one critical in the XML parser reachable from POST /v1/import, two medium findings in the internal reconciliation job (both already have named owners and expiries), and four low findings in test fixtures. The team can disable POST /v1/import in the production config, three internal clients would need two weeks to migrate, and can deploy a WAF rule that blocks XML content types. The parser fix lands in three weeks. What should the release decision be?
MCQ-9A platform team gates releases on a SAST scan. The scan covers 12 of the company's 240 services: the 12 that handle payments. `baseline.json`, committed in April 2023, lists 9,412 findings as accepted with no expiry. After a noisy rollout in January, three rules were moved to `disabled_rules`: the ones that flag SQL-injection sinks, hardcoded credentials, and unsafe deserialization. Last quarter the gate blocked 0 of 63 releases. Release 58, in the payments service, shipped a SQL injection on a line added that quarter. Which single change would have made the gate catch it?
MCQ-25A payments platform holds its Q3 secure-SDLC review. Q3 figures, with Q2 in parentheses: scanner jobs 42,000 (32,000); repositories with SAST and SCA enabled 96% (71%); findings closed 3,100 (2,600); findings opened 1,180 (1,050); mean time to remediate critical findings 61 days (34 days); production incidents 3 (1). Two of the three incidents traced to dependencies carrying critical CVEs that the scanners had flagged a median of 5 months before the incident, and neither CVE had been assigned to a team. The summary slide reads: "Coverage and throughput both up. Program trending well." Which reading of the quarter is correct?
MCQ-22A payments company has 40 product teams and an AppSec team of 3 that also carries pentest triage and incident support. Each team has a security champion at 20% time; champions write the threat model for every new service, and AppSec reviews and signs off on each one before the design can ship. Champions produce 80 threat models a quarter, and AppSec spends about 6 hours on each review. The review queue now holds 30 models, and median design-to-ship security time is 9 days: the same as before the program existed. Champions say the work is worth their time. Which change addresses the bottleneck the numbers show?
MCQ-23A 120-engineer payments company runs an OWASP SAMM assessment each year. This year's scores are Governance 1, Design 3, Implementation 3, Verification 3, Operations 1. Over the same 12 months there were 11 Sev-1 production incidents; 9 were traced to third-party components with published CVEs that were already in the dependency tree when the service shipped. The team can fund exactly one new control this quarter. Which one should it fund?
MCQ-20A service due to ship in two weeks returns 340 SAST and dependency findings, and the team cannot fix them all before the date. The service is internet-facing and parses untrusted input on several endpoints. Which triage approach should set the remediation order?
MCQ-24A vendor is retiring firmware 4.1 for an on-premise security gateway in 90 days. The firmware verifies update packages with an RSA-2048 key whose private half sits in the vendor's HSM. Telemetry shows 300 customers still run 4.1; 80 of them use a hardware revision that cannot run 5.0 and have no upgrade path. The vendor will publish one final signed 4.1 update that points the update client at a read-only archive. Which action belongs in the retirement plan so those 80 customers can still receive vulnerability notices without leaving a key that can push code to every 4.1 device?