Domain 3 Practice Test: Security Architecture and Engineering

How to take this test

This test covers everything Domain 3 taught in Chapters 12 through 14: the design principles and the trust boundary, the reference monitor and the architecture models, Zero Trust as NIST SP 800-207 frames it, the cryptographic primitives and their standards, the key lifecycle, the formal security models from Bell-LaPadula to Brewer-Nash, the evaluation industry of Common Criteria and FIPS 140, and the hardware and physical layers underneath it all. Twenty-five questions cannot sample a domain this wide the way the adaptive exam does, but each question here is built on a decision pattern the real exam actually scores, so a wrong answer tells you which pattern you are missing rather than which fact you forgot.

Set the clock to thirty minutes before you read the first stem: a minute and a quarter per item, the pace this book assumes for a full 150-item session. Answer every question, including the ones you would rather skip, because a blank is wrong and the engine never asks why you submitted what you did. Do not look at the answer key early, and do not return to a question once you have moved past it. The real exam offers no item review, and the practice that transfers is the practice that reproduces that constraint. When a question makes you guess, guess, and keep moving; the score sheet will do the analysis afterward.

One habit matters more than any vocabulary list in this domain. Before you choose, name the discipline the stem is really testing: is this a principle question, a crypto question, a model question, or a hardware and physical question? Domain 3 is four subjects wearing one name, and most stems here are written so that two options are technically familiar and only one fits the facts as stated. In the crypto questions, remember the Chapter 13 rule: the algorithm is the cheap part and the key is the expensive part, so crypto answers are usually about key management, randomness, and lifecycle rather than cipher arithmetic. In the model questions, read the levels: who is the subject, what is the object, and which direction does the rule forbid? When you are still torn, apply the Chapter 2 tie-breaker: which option would a responsible manager defend in front of the board, with the standard named and the decision documented?

When you are done, score yourself honestly and read the score interpretation before you touch anything else. The number is a triage instrument, not a verdict. It tells you which of the three preceding chapters deserves your next study hours, and that is worth more than the number itself.

The 25 questions

1. A bank plans to add a badge reader to the door of the room that holds its core processing systems. The discussion turns to what the door should do when power fails. Which decision is correct?

   A. The door must fail open so staff are never locked out during an outage B. The door must fail to the locked state, because a security door that unlocks on failure hands an attacker the room at the worst possible moment, and life-safety exits are the doors that fail open for evacuation C. The door should remain in whatever state it was in when power failed D. The door should fail open during business hours and locked outside them

2. A company layers controls on its customer portal: a web application firewall at the edge, input validation in the application, parameterized queries at the data layer, and encryption in transit and at rest, so that a failure in any single layer does not expose the data. What principle does this arrangement embody?

   A. Separation of duties B. Defense in depth C. Economy of mechanism D. Need to know

3. A security kernel is designed so that every access request, without exception, passes through the enforcement mechanism before any object is touched, the mechanism itself runs in a way that no application can modify, and the mechanism is small enough that its correctness can be verified. The requirement that every access be mediated is known as:

   A. Complete mediation B. Least privilege C. Non-interference D. The Chinese Wall

4. An audit finds that the nightly backup service account holds domain administrator rights, granted years ago so the backups would never break. The backup product needs only read access to the servers it protects. What is the correct response?

   A. Keep the rights, because a broken backup is worse than the risk B. Replace the account with credentials scoped to only the access the backup function needs C. Keep the account but require a password change every month D. Move the backups to a different vendor’s product

5. In the Zero Trust architecture of NIST SP 800-207, a request arrives for a resource. One component evaluates the request against policy and against signals such as device health, user identity, and observed behavior, and determines whether access is permitted. A second component carries that determination to the enforcement point, which allows or denies the connection. The component that makes the permit or deny determination is the:

   A. Policy enforcement point B. Policy administrator C. Policy engine D. Data access policy

6. A security vendor markets its encryption product by insisting the algorithm stay secret, arguing that hiding the algorithm protects its customers. Which design principle does this marketing invert?

   A. Economy of mechanism B. Open design C. Fail-safe defaults D. Least common mechanism

7. A government system stores data at Top Secret, Secret, and Confidential levels. Users log on at the classification level they are cleared for and can then access data at or below that level; no user may access data above their clearance. Which security mode describes this system?

   A. Dedicated B. System high C. Compartmented D. Multilevel

8. In the finance department, the employee who creates a vendor account may not also approve payments to that vendor. The two steps are assigned to different people so that fraud requires collusion. Which principle is this?

   A. Least privilege B. Separation of duties C. Need to know D. Psychological acceptability

9. An analyst encrypts a set of archived reports with a modern block cipher and stores the ciphertext. Later, an attacker who cannot read the ciphertext modifies it in storage by flipping bytes. Which statement about the protection is correct?

   A. The encryption proves the reports were not altered, because the attacker could not decrypt them B. Encryption alone provides confidentiality, not integrity, and modified ciphertext can decrypt to modified plaintext unless an integrity mechanism is used C. Modification is impossible because block ciphers detect tampering D. The analyst should have used a longer key to prevent modification

10. A standards review finds legacy systems using DES for at-rest encryption. Which replacement is the current FIPS-approved symmetric standard?

    A. AES with 128, 192, or 256-bit keys, defined in FIPS 197 B. 3DES as defined in SP 800-67 C. Blowfish with a 448-bit key D. RC4 in stream mode

11. A developer encrypts a structured document with a block cipher in ECB mode. A reviewer sees that repeated plaintext blocks produced identical ciphertext blocks. What is the reviewer’s correct conclusion?

    A. ECB is fine because the cipher is strong B. ECB leaks plaintext structure, because identical blocks encrypt identically and patterns survive; a mode such as CBC with a unique IV, or GCM, should be used C. The problem is the key, not the mode D. ECB is secure once the document is compressed

12. A company’s directory stores password verifiers as salted digests computed with a deliberately slow key derivation function such as PBKDF2, the function NIST SP 800-132 describes. Why salt, and why slow?

    A. The salt makes the digest unreadable, and speed makes the database faster B. The salt defeats precomputed tables such as rainbow tables, and the work factor raises the cost of each offline guess, slowing brute force after a database theft C. The salt is the password, and the slow function is an implementation detail D. Salting is only required for government systems

13. Two internal services share a secret key and authenticate their messages with a keyed hash per FIPS 198-1. What does a valid message authentication code prove, and what does it not prove?

    A. It proves the message came from the holder of the shared key, so it provides integrity and data-origin authentication, but it does not provide non-repudiation, because either side could have produced it B. It proves non-repudiation, because the key is secret C. It proves the message was encrypted D. It proves only that the message was received recently

14. A legal agreement is digitally signed with the signer’s private key, and the signature verifies against the signer’s public key certificate. Which property does the valid signature provide that a shared-key mechanism could not?

    A. Confidentiality of the agreement B. Non-repudiation, because only the holder of the private key could have produced the signature, and the certificate binds that key to the signer C. Availability of the agreement D. Faster verification than any other mechanism

15. A relying party must check, at the moment of a transaction, whether a presented certificate is still valid, and the status must come back in real time rather than from a list refreshed on a schedule. Which mechanism answers this need?

    A. A certificate revocation list downloaded hourly, per RFC 5280 B. The Online Certificate Status Protocol (OCSP), per RFC 6960 C. A certificate pinning list compiled once a year D. A fresh certificate signing request

16. An operations team has kept a bulk-encryption key in service for years past the cryptoperiod set in its key management plan, because rotation is disruptive. Per NIST SP 800-57, why is this decision dangerous?

    A. The key will expire and stop working without warning B. The cryptoperiod limits how long a key is used and how much data is protected under it, so overage increases the data exposed to a single compromise and lengthens the cryptanalysis window C. Cryptoperiods exist only for asymmetric keys D. A key used longer consumes more entropy

17. In a system enforcing Bell-LaPadula, a user holds a session at the Secret level. The system holds a Top Secret document and a Confidential document. Which action does Bell-LaPadula forbid for this user?

    A. Reading the Top Secret document B. Reading the Confidential document C. Writing to the Top Secret document D. Writing to the Secret document

18. An embedded device enforces a strict integrity policy: a low-integrity software component attempts to overwrite a high-integrity configuration file, and the operating system denies the write. Which model is this behavior from?

    A. Bell-LaPadula B. Biba C. Brewer-Nash D. The Common Criteria

19. A core banking system is designed so that every transaction must follow a certified transformation procedure, every change to a constrained data item is logged, and no single person can both create and approve a transaction. Which security model does this design implement?

    A. Bell-LaPadula B. Biba C. Clark-Wilson D. Brewer-Nash

20. A consulting firm that advises two competitors in the same industry has a policy that a consultant who has accessed one client’s internal data may never access the other client’s data. Which model formalizes this rule?

    A. The Chinese Wall model of Brewer-Nash, which groups data into conflict-of-interest classes B. Bell-LaPadula C. Biba D. The Graham-Denning model

21. A procurement team writes a reusable specification of the security requirements for a class of products, say, firewalls for a banking environment, so that any vendor can evaluate its product against the same requirements. In Common Criteria terms, per ISO/IEC 15408, this artifact is a:

    A. Security Target B. Protection Profile C. Target of Evaluation D. Evaluation Assurance Level

22. A vendor markets a cryptographic module validated at FIPS 140-3 Level 4, the standard harmonized with ISO/IEC 19790. What capability does Level 4 provide beyond Level 3?

    A. The module erases or renders itself inoperable when tampering is detected or attempted, and it is protected against environmental variations, where Level 3 provides tamper evidence and identity-based operator authentication B. The module is tamper-evident, which is all any level requires C. The module runs faster D. The module never requires key management

23. A laptop’s UEFI firmware and bootloader are measured at boot, the measurements are stored in the platform configuration registers of the machine’s TPM, and a corporate service compares them against known-good values before the laptop may reach the network. What is this verifying?

    A. The identity of the user typing the password B. The integrity of the boot chain, that the machine booted a known-good configuration and was not started by untrusted code C. The absence of physical damage to the laptop D. The laptop’s network address

24. An organization generates a new master key on its hardware security module. The ceremony requires three named officers to be present, each holding a share of the key material, and a video record is kept. Which controls are in effect?

    A. Split knowledge and dual control, so no single insider can reconstruct or use the key alone B. Least privilege and need to know C. Fail-safe defaults D. Separation of duties only, with no technical control

25. A 24/7 data center houses live electronic equipment, and the fire suppression design must extinguish a fire without destroying the equipment it protects. Which suppression approach fits best?

    A. Standard water sprinklers B. A clean agent gas system per NFPA 2001, with agents such as FM-200, Novec 1230, or Inergen C. Halon, which remains the best agent for new installations D. Dry chemical extinguishers mounted on the walls

Answer key

Rationales

1. B. A door that guards a security zone should fail to the secure state, locked, so that a power failure does not unlock the room for anyone who is waiting for the lights to drop. The doors that fail open are life-safety exits, whose job is to let people out during an evacuation, and the two jobs should never be assigned to the same design. Options A, C, and D each leave the room unlocked or open at some point the attacker can predict.

2. B. Defense in depth means placing multiple independent controls so that no single failure exposes the asset: the firewall, the input validation, the parameterized queries, and the encryption each covers a layer the others do not. Separation of duties splits one sensitive action among people (option A), economy of mechanism favors simple designs (option C), and need to know restricts access to what a role requires (option D).

3. A. The reference monitor concept, which the TCSEC formalized, rests on three properties: complete mediation, meaning every access is checked; isolation, meaning the monitor cannot be modified by subjects; and verifiability, meaning the monitor is small enough to prove correct. Mediating every request is complete mediation. Least privilege (option B) is about limiting rights, non-interference (option C) is an information-flow property, and the Chinese Wall (option D) is Brewer-Nash’s conflict-of-interest rule.

4. B. Least privilege, the control behind NIST SP 800-53 AC-6, requires that each account hold only the permissions its function needs. The backup service needs read access to the protected servers, not domain administration, so the account should be re-scoped and re-credentialed. Keeping the rights for convenience (option A), changing the password on an overprivileged account (option C), or changing vendors (option D) all leave the actual defect, excessive privilege, in place.

5. C. NIST SP 800-207 names three logical components: the policy engine, which makes the access decision from policy and its inputs; the policy administrator, which relays the decision and provisions or revokes the path; and the policy enforcement point, which executes the allow or deny. The engine decides, so option C is correct. Options A and B are the executing and relaying components, and option D is a policy artifact, not a component.

6. B. Open design, one of Saltzer and Schroeder’s principles, holds that security must not depend on keeping the design or implementation secret; the secrecy that matters is the key material. This mirrors Kerckhoffs’s principle, that a cryptosystem should remain secure even if everything about it except the key is public. Hiding the algorithm is security by obscurity. Options A, C, and D are real principles, but none of them says what this marketing violates.

7. D. The TCSEC security modes describe who may use a system and what they may see. Multilevel mode fits exactly: the system processes data at multiple classification levels, and users at different clearances log on and access data at or below their own level. Dedicated mode (option A) requires every user to be cleared and have need to know for everything, system high (option B) requires every user to be cleared for the highest level processed, and compartmented mode (option C) applies the same clearance to all users with compartments gating access.

8. B. Separation of duties splits a sensitive operation into steps assigned to different people so that completing the operation requires collusion, the control mapped in NIST SP 800-53 AC-5 and ISO/IEC 27001:2022 Annex A control 5.3. Least privilege (option A) limits the extent of rights, need to know (option C) limits access to data by role, and psychological acceptability (option D) is about usability of controls.

9. B. Encryption alone provides confidentiality. It does not provide integrity: an attacker who can flip bits in ciphertext will, with many modes, cause a corresponding change in the decrypted plaintext. Integrity needs a separate mechanism, a MAC or hash, and authenticated modes such as GCM per NIST SP 800-38D bundle both properties into one operation. Options A, C, and D each attribute a property to plain encryption that it does not have.

10. A. AES with 128, 192, or 256-bit keys, defined in FIPS 197, is the current approved symmetric cipher. DES is withdrawn, and 3DES, defined in SP 800-67, is disallowed for new use under NIST SP 800-131A. Blowfish (option C) and RC4 (option D) were never FIPS-approved standards.

11. B. In ECB mode, the mode NIST SP 800-38A defines, each plaintext block is encrypted independently, so identical plaintext blocks produce identical ciphertext blocks and patterns in the data survive in the ciphertext. The fix is a chaining or authenticated mode such as CBC with a unique IV, or GCM. The cipher’s strength (option A), the key (option C), and compression (option D) do not cure the mode’s structural leak.

12. B. A salt makes each stored digest unique even when many users choose the same password, which defeats precomputed tables such as rainbow tables, and a deliberately slow function with a work factor raises the cost of each candidate guess, which slows offline brute force after the digest database is stolen. The salt does not make the digest unreadable in the sense option A implies, and salting and work factors are standard practice for any system that stores verifiers.

13. A. An HMAC computed with a shared secret, per FIPS 198-1, provides integrity, the message was not altered, and data-origin authentication, the message came from whoever holds the key. It cannot provide non-repudiation: because both parties share the same key, either one could have produced any given tag, so a third party cannot tell them apart. Options B, C, and D claim properties the mechanism does not deliver.

14. B. A digital signature binds a message to a private key, and the certificate, per RFC 5280, binds that key to the signer’s identity. Because only the holder of the private key can produce a valid signature, the signer cannot plausibly deny having signed, which is non-repudiation, and a shared-key mechanism cannot give that property because either party holds the key. Confidentiality (option A), availability (option C), and speed (option D) are not what signatures provide.

15. B. OCSP, defined in RFC 6960, answers a single certificate’s status in real time with a response of good, revoked, or unknown, which is what a relying party needs at the moment of a transaction. A CRL, defined in RFC 5280, is a periodically published list with the freshness gap that implies. Pinning (option C) and certificate signing requests (option D) do not answer revocation status at all.

16. B. NIST SP 800-57 defines the cryptoperiod as the time span during which a key is authorized for use, a limit driven by the amount of data protected, the exposure that grows with use, and the window an attacker has to collect ciphertext for cryptanalysis. Running past it concentrates more data and more time under one key. Keys do not expire and stop working on a calendar (option A), the concept applies to symmetric keys above all (option C), and entropy is not consumed by use in the way option D suggests.

17. A. Bell-LaPadula enforces confidentiality with two rules: the simple security property, no read up, and the star property, no write down. A Secret-level subject may not read the Top Secret document, because reading up violates the simple security property. Reading down to Confidential is allowed, and writing up, to Top Secret, is permitted under the star property’s logic, as is writing at the same level. Only option A is forbidden.

18. B. Biba is the integrity mirror of Bell-LaPadula: its star integrity property forbids a low-integrity subject from writing to a high-integrity object, and its simple integrity property forbids reading down from a low-integrity subject. The denial in the scenario is exactly the no-write-up rule. Bell-LaPadula (option A) protects confidentiality, Brewer-Nash (option C) protects against conflicts of interest, and the Common Criteria (option D) is an evaluation framework, not a policy model.

19. C. Clark-Wilson is the commercial integrity model: well-formed transactions, meaning operations move only through certified transformation procedures, constrained data items whose changes are logged and validated, and separation of duties so that no single person can both make and approve a change. Bell-LaPadula (option A) and Biba (option B) are lattice policies about reading and writing between levels, and Brewer-Nash (option D) is about conflicts of interest.

20. A. Brewer-Nash’s Chinese Wall model is built for conflict of interest: data is grouped into conflict-of-interest classes, and a subject who has accessed data in one company’s dataset may not access data from a competing company in the same class. The consultant scenario is the model’s canonical use. The other options enforce confidentiality, integrity, or administration rules that do not express the wall.

21. B. A Protection Profile is a reusable statement of consumer security requirements for a class of products, independent of any specific implementation, which is exactly what a procurement team writes once for many vendors. A Security Target (option A) is a vendor’s claim about a specific product, the Target of Evaluation (option C) is the product being evaluated, and the Evaluation Assurance Level (option D) is the depth of the evaluation, EAL 1 through 7 in ISO/IEC 15408.

22. A. FIPS 140-3, harmonized with ISO/IEC 19790, sets four increasing levels. Level 3 adds tamper evidence and response plus identity-based operator authentication, and Level 4 adds tamper resistance and protection against environmental variations, with the module expected to erase or render itself inoperable under attack. Option B describes only Level 2, and options C and D describe properties the standard does not rate.

23. B. A TPM, the trusted platform module of ISO/IEC 11889, stores boot measurements in platform configuration registers and can report them to a verifier, the remote attestation flow. Comparing the measurements against known-good values verifies that the boot chain ran untrusted code nowhere, that the firmware and bootloader are what the organization approves. The user’s password (option A), the laptop’s physical condition (option C), and its address (option D) are not what the measurement covers.

24. A. The key ceremony uses split knowledge, no single person holds the full key material, and dual control, the key is generated and used only when the required number of officers act together, with the video record providing the audit trail. Together they defeat the insider who would otherwise reconstruct or use the key alone. Options B and C are other principles, and option D misses that the ceremony is itself the technical control.

25. B. A clean agent gas suppression system per NFPA 2001, using agents such as FM-200, Novec 1230, or Inergen, extinguishes fire without leaving residue and without damaging live electronics, which is why it fits data centers. Halon was the old leader, but its production was banned under the Montreal Protocol, so option C is wrong for new installations. Water (option A) and dry chemical (option D) both destroy the equipment the room exists to protect.

Reading your Domain 3 score

Score yourself against the bands Chapter 1 set for every domain test in this book:

The single number hides the information you need, so break the misses down by area. Domain 3 is four disciplines under one label, and the triage that matters is at the topic level:

Two habits make the score useful. First, log the result and the area misses on your score sheet, because the full practice exam in Chapter 33 will re-test this domain and you want the comparison. Second, treat a miss pattern as a question about process, not just facts. If your misses cluster on questions where you picked a technically familiar option that did not fit the scenario, your problem is stem reading, and the fix is to name the discipline, principle, crypto property, model, or control, before you look at the options. If the misses cluster on questions with named standards, FIPS 197, RFC 6960, ISO/IEC 15408, your problem is source recall, and the fix is to redraw each standard’s structure cold, without the chapter open.

What the score means for your plan

Domain 3 carries 13 percent of the current outline’s weight, tied with communication and network security and identity and access management for the second-largest share behind security and risk management. It is also the domain that pays for itself twice. The cryptography you just tested is the foundation of TLS and VPN security in Domain 4. The least privilege, separation of duties, and access decisions you just reasoned through are the vocabulary of identity and access management in Domain 5. The physical controls appear again when security operations deals with facility protection in Domain 7. A candidate who treats Domain 3 as one exam section to memorize and forget is discarding the reasoning engine the rest of the book runs on.

Pass this test at the level your score band demands and you have earned the right to move to Part V with confidence. Fail it, and you have earned something rarer: a precise list of what to re-read, organized by the discipline that matters. Either way, the score sheet is now the authority on where your Domain 3 hours go. The next chapter opens Domain 4, Communication and Network Security, where the trust boundaries and the cryptographic primitives you just worked with get wired together into networks, segmentation, and secure channels.