Overview
Curated: · Written: · Reviewed:
Compliance is scoped assurance, not a guarantee of security
Interviewers use compliance questions to test one judgement: can you tell the difference between an artifact that proves something bounded and a logo that proves nothing? A weak answer recites framework names and says "we're SOC 2 certified." A strong answer names the artifact, its scope, its period, and what it does not cover. This guide is organized around that judgement.
What interviewers probe, and what a weak answer sounds like
The recurring questions:
- "We're SOC 2, ISO 27001, and PCI — what does that actually mean?" They want you to say attestation vs. certification, scope, and period without being asked.
- "A customer asks if you're GDPR certified. What do you say?" GDPR is law, not a security certification — and although Articles 42–43 do establish voluntary data-protection certification mechanisms and seals, those cover specific processing operations, not an organization's overall security, and they are not a general compliance proof. The trap is whether you know that and whether you defer legal interpretation to counsel while owning the technical evidence.
- "Your SOC 2 report has an exception on change management. Walk me through the risk." They want root cause, affected period, compensating controls, and whether the exception is isolated or population-wide.
- "Sales wants to put the ISO logo on the deck. The certificate covers the HQ ISMS only. Go." They want you to refuse the claim and explain what a truthful claim looks like.
A weak answer sounds like: "SOC 2 proves we're secure," "ISO 27001 is the international standard so we're compliant everywhere," or "PCI means we don't store card numbers." Each collapses a scoped, time-bound conclusion into a guarantee. If you catch yourself saying "certified" about SOC 2, stop — SOC 2 is an attestation, not a certification.
Likely follow-ups: who audits each instrument, what artifact the customer actually receives, what happens when scope changes mid-period, and how you run one control set against five frameworks without five teams.
The instrument map: what each one actually is
SOC 2 (AICPA Trust Services Criteria) is an attestation: a CPA firm issues an opinion on controls at a service organization. Type I covers design at a point in time; Type II covers operating effectiveness over a period — that period, typically 3–12 months, is the whole value of the report. Criteria are selected: Security is mandatory; Availability, Processing Integrity, Confidentiality, and Privacy are in scope only if included. The report includes a system description, the auditor's opinion, exceptions, complementary user-entity controls (CUECs), and subservice-organization treatment (carve-out vs. inclusive). A logo or clean summary page is not the report; the report is the report.
ISO/IEC 27001:2022 specifies requirements for an information security management system: context, scope, leadership, risk assessment and treatment, competence, operation, performance evaluation, continual improvement. Certification is performed by an accredited certification body against a defined ISMS scope — which can be one product, one entity, or the whole company. The Statement of Applicability (SoA) justifies which Annex A controls are included and excluded. ISO/IEC 27002 is the companion guidance: implementation advice for those controls, not a certifiable standard. A certificate whose scope reads "HQ ISMS" says nothing about your us-east-1 production environment.
PCI DSS v4.0.1 is a prescriptive baseline for entities that store, process, or transmit cardholder data or sensitive authentication data, or can affect the cardholder data environment. Artifacts: an Attestation of Compliance (AoC), and for larger merchants a Report on Compliance (ROC) from a Qualified Security Assessor; smaller merchants may self-assess via an SAQ. Scope is driven by accurate data-flow diagrams and segmentation — tokenization or outsourcing shrinks scope only when the architecture genuinely removes systems from the CDE.
NIST CSF 2.0 is a voluntary outcome taxonomy — Govern, Identify, Protect, Detect, Respond, Recover. No certification exists. NIST SP 800-53 is a control catalog organizations tailor through a risk-management process; adopting its identifiers without parameters, implementation, and assessment is not a control program. FedRAMP applies defined baselines (built on 800-53), documentation, third-party assessment, agency authorization, and continuous monitoring to cloud services used by US federal agencies. CMMC adds assessed maturity levels for defense-industry contractors handling controlled unclassified information (CUI).
HIPAA is US law covering covered entities and business associates handling protected health information — a Security Rule with administrative, physical, and technical safeguards, not a certification. There is no "HIPAA certification"; there are audits, risk assessments, and attestation of compliance.
GDPR is law, not a security certification. Applicability, controller/processor roles, lawful basis, purpose limitation, minimization, data-subject rights, records of processing, data protection by design, incident obligations, and transfer mechanisms need qualified legal and privacy interpretation. Articles 42–43 do provide for voluntary data-protection certification mechanisms and seals (accredited bodies, approved criteria — e.g., Europrivacy as an EU-approved scheme), but these certify conformity of specific processing operations or products against approved criteria; they are not a general security certification and most organizations operate without one. Engineers supply accurate data flows, safeguards, and evidence; they do not invent legal conclusions.
Which one does a given business need?
| Situation | What customers ask for | Why |
|---|---|---|
| SaaS selling to US enterprises | SOC 2 Type II, Security (often + Availability, Confidentiality) | Enterprise vendor-risk reviews are built around reading SOC 2 reports |
| SaaS selling into EU/global | ISO 27001 certification | Recognized across jurisdictions; pairs with GDPR DPA and transfer mechanisms |
| Handling card data | PCI DSS AoC (or ROC) | Contractual obligation from card brands and acquirers |
| US federal agency customer | FedRAMP authorization | Agencies generally cannot buy unauthorized cloud services |
| Defense contractor with CUI | CMMC at the required level | Contract requirement |
| Handling PHI | HIPAA compliance program + BAA | Legal obligation, evidenced by risk assessment and controls, not a certificate |
These stack: a payments SaaS serving enterprises commonly runs SOC 2 + ISO 27001 + PCI DSS simultaneously, which is why unified control sets matter (below).
Compliance vs. security: what passing proves and doesn't
Passing an assessment supports a bounded conclusion: for this scope, these criteria, this period, with this evidence, the controls operated as described. It does not prove absence of breaches, defects, or untested risks. The audit samples a population over a window; anything outside the sample or the window is untested.
Where checkbox compliance leaves residual risk:
- Controls that satisfy the criterion but not the threat. A quarterly access review passes; the compromised key created yesterday is still live.
- Scope boundaries drawn for convenience. The audited system excludes the staging environment that holds a copy of production data.
- Evidence that measures paperwork, not behavior. Policy documents exist; nobody follows them.
- Point-in-time snapshots presented as continuous operation.
The architect's job in an interview is to argue for controls the framework does not force: "SOC 2 doesn't require runtime detection on the tokenization vault, but here's the threat model and here's what I'd add." That answer — framework as floor, risk model as driver — is what separates senior candidates.
Attestation vs. certification vs. self-assessment
| Who assesses | Artifact | What a customer accepts | |
|---|---|---|---|
| SOC 2 Type I | CPA firm | Report: design at a point in time | Rarely sufficient on its own for vendor risk |
| SOC 2 Type II | CPA firm | Report: opinion on operating effectiveness over a period | The standard enterprise vendor-review artifact |
| ISO 27001 | Accredited certification body | Certificate + scope statement + SoA (surveillance annually, recertification ~3 years) | The certificate and its scope statement |
| PCI DSS | QSA (or self-assessment for eligible merchants) | AoC, ROC for larger merchants | AoC with matching scope and data-flow diagrams |
| CMMC | C3PAO assessors | Assessment at a level against the applicable model | Level-specific certificate in SPRS |
| NIST CSF / 800-53 self-assessment | The organization | Profile, control implementation, POA&M | Depends entirely on the receiver's trust in your process |
The pattern: the further the artifact gets from an independent assessor examining operating effectiveness over time, the more the customer is really evaluating you rather than the artifact.
Operate one control system
Build an obligation register with source/version, applicability rationale, interpretation owner, and effective date. Normalize requirements into a control catalog without erasing framework-specific language. Each control carries: objective, owner, operator, systems/data, procedure or policy, frequency or event trigger, evidence, dependencies, test, exception, and mapped obligations.
Crosswalks reduce duplicate work — one access-review control can satisfy ISO 27001 A.5.18, a SOC 2 CC6 logical-access criterion, and PCI DSS Requirement 7 — but mappings are many-to-many, directional, and context-dependent. One framework's evidence may not satisfy another's scope or assessment method: a SOC 2 test over 6 months does not automatically cover a PCI requirement tested differently over a different population. Where frameworks genuinely diverge — SOC 2 and CSF are outcome-based, PCI DSS and 800-53 are prescriptive — the prescriptive requirement sets the implementation floor and the outcome-based criterion accepts it as one valid implementation.
Scoping: the decision that makes or breaks the audit
Scope must be reproducible from authoritative inventories and data flows, not chosen as a convenient diagram boundary. Record legal entities, products and versions, deployment models, locations, people, technology, data categories, suppliers, interfaces, and the assessment period — then reconcile the declared population against cloud inventory, identity systems, repositories, network observations, procurement, and processing records.
Where each instrument's scoping lives:
- SOC 2: which Trust Services Criteria you include, and the system description that bounds the audited system. Adding Availability to the criteria means adding uptime evidence; drawing the boundary to exclude a customer-facing dependency invites a qualification.
- ISO 27001: the scope statement plus the SoA. Excluding a business unit is legitimate if risk treatment justifies it — and the certificate then says so.
- PCI DSS: data-flow-driven. Every system that stores, processes, or transmits PAN/SAD, or connects to one, is in scope until segmentation is validated.
A badly drawn boundary breaks the audit in predictable ways: the auditor samples a system you forgot was in scope, the system description doesn't match reality, or a mid-period architecture change (new region, new subprocessor, acquisition) silently invalidates the declared population. Reassess applicability and external claims when scope changes, before release — not at the next annual audit.
Segmentation, tokenization, outsourcing, and managed services reduce obligations only when the technical boundary is effective and the retained responsibility is explicit. A provider's report does not cover the customer's tenant configuration, identities, data use, or complementary controls unless the instrument says so. Inherited controls need a consumer-specific contract: provider, objective, covered component, configuration assumptions, evidence, failure notification, fallback. Test that the consuming product actually uses the inherited service as declared, and identify shared dependencies that make several controls fail together.
Assessment, findings, and truthful claims
An assessment plan ties each criterion to population, procedure, sample rationale, evidence period, and evaluator independence. Automated evidence improves coverage, but collectors, queries, identities, clocks, and retention are part of the evidence chain and must themselves be governed. Investigate failures outside the sample: isolated exception, population-wide design defect, or inaccurate scope? Management responses state root cause, affected period, compensating action, owner, and verified remediation. Deleting the failed record or rerunning until green destroys the assurance value.
Evidence is authoritative, attributable, time-bound, and reproducible: configuration and query output, tickets and approvals, logs, sampled records, test results, training records, incident and recovery artifacts. A screenshot is weak when a versioned export or automated query can show population and time. Assess both design effectiveness — could the control meet its objective? — and operating effectiveness — did it operate across the stated population and period?
Exceptions state affected requirement/control, scope, rationale, residual risk, compensating measures, owner, approver, start/expiry, monitoring, and closure evidence. Continuous monitoring catches drift, expired evidence, failed control runs, supplier and architecture change, new obligations, and incidents. Metrics emphasize population coverage, control-test outcomes, exception age, recurrence, and actual risk reduction — not documents produced or controls mapped.
Public and customer claims must match the exact artifact: entity, service, system boundary, criteria, report type, period, certificate status, exclusions, exceptions, and customer responsibilities. Don't turn a framework mapping into certification, a point-in-time test into continuous operation, or an attestation into regulator approval and breach immunity. Give claims an owner and expiry, monitor published wording, and correct material inaccuracies.
Worked example: a SOC 2 Type II is not a PCI + GDPR badge
Vendor pack for a payments SaaS. Sales wants "we're certified."
| artifact | actual scope | customer claim that holds |
|---|---|---|
| SOC 2 Type II, Security TSC, 2025-01-01–2025-06-30, AWS carved out | that period, those criteria, that system description | not "PCI certified" |
| ISO/IEC 27001 certificate, ISMS = HQ building | HQ ISMS | not us-east-1 production |
| PCI AOC: PAN tokenized, CDE = tokenizer + vault | cardholder environment only | not GDPR lawful basis or EU transfers |
The interview answer is the system description and the period. A mapping spreadsheet is not an attestation. If the interviewer follows up with "so what would you tell sales?" — the truthful claim names each artifact, its scope, and its period, and gives the claim an owner who re-checks it when the next report period ends.
