T Techclick ← All lessons
ISC2 · CISSP Domain 3 · Interactive lesson

CISSP Domain 3 — Build security in, not on

A vendor-payment API shipped wide-open. One engineer both created the vendor and approved the wire. Overnight a forged payment left the ledger. The board asks why “the WAF” did not stop it. Domain 3 is the 13% of the CISSP exam where you answer as a manager and as an engineer — the principle, the model, the crypto, and the room — before anyone buys another box.

22 min read · L2 primary · manager + engineer · Quiz at end

⚡ Quick Answer

CISSP Domain 3 (13%): pick the design principle, the security model, the crypto family, and the site control — as a manager who scopes risk and an engineer who proves the control.

After this page you can

Quick answer

Domain 3 is not “memorise BLP arrows.” It is name the property the system must enforce, then bake the control into design, then prove it after go-live. Simple = read. Star (*) = write. Bell-LaPadula keeps secrets: no read up, no write down. Biba keeps integrity: no read down, no write up. Clark-Wilson adds well-formed transactions and separation of duties — a lattice alone is not a ledger. Brewer-Nash (Chinese Wall) is history-based conflict of interest. Bulk data uses a symmetric cipher; the session key rides an asymmetric (or PQC) wrap. Occupied server rooms get a clean agent, not water and not CO2. A manager’s first move after a forged payment is the requirement and the model, not a purchase order for a WAF.

Say this out loud

I do not bolt security onto a finished box. I pick the property — secrecy, integrity, conflict, or safety — then the principle, then the crypto or the room, and I leave evidence that the control fails closed.

1. Why Domain 3 matters

ISC2 weights Security Architecture and Engineering at 13% of the CISSP CAT (exam outline effective 15 April 2024). The official verbs run from “research, implement and manage engineering processes using secure design principles” through models, system capabilities, crypto, site design, and the information-system lifecycle. The exam — and a real architecture board — punish the same mistake: treating security as a product you add after the code ships.

That is the ticket in this lesson. The API defaulted to open. One person owned create and approve. The WAF never saw a well-formed but fraudulent transformation. Domain 3 would have stopped that at design time: secure defaults, segregation of duties, and a Clark-Wilson access triple — subject, approved program, constrained data item.

Hero · design, model, crypto, site
Layered building, classification lattice, crypto lock, and site rings — security is designed in
Notice: four jobs, one domain. Principles first. Then the formal rule. Then the key. Then the room. A WAF is none of those by itself.

Manager lens

Which property must this system enforce, who owns residual risk, and which document is mandatory? Buy nothing until the requirement and the fail-closed behaviour exist on paper.

Engineer lens

Which rule, algorithm, or physical control actually implements that property? If the control cannot fail closed or the private key is shared, the model is a slide.

Official outline (what this page covers)

3.1 Secure design principles (threat modeling through SASE). 3.2 Security models (Biba, Star Model, Bell-LaPadula). 3.3–3.5 Select controls; IS capabilities (memory protection, TPM, encryption); architecture vulnerabilities. 3.6–3.7 Cryptographic solutions and cryptanalytic attacks. 3.8–3.9 Site and facility. 3.10 Information-system lifecycle. Source: ISC2 CISSP Certification Exam Outline. Domain 3’s official weight is 13% of the whole exam — that is the only percentage this page uses.

2. Mental model — principles and models

Hold three parts. Interviews fail when people treat CIA as a slogan, treat “policy” as any PDF, or recite BLP arrows without saying which property the stem needs.

Journey · goal, lattice, rule, enforce
Four glass panels labeled Goal, Lattice, Rule, and Enforce
Read left to right. Enforce is not optional decoration. If the reference monitor is not always invoked, the lattice is theatre.

Part 1 · hard words before the arrows

Pre-train these before the runbook

Subject / object — who acts versus what is acted on. A process is a subject; a file is an object.

Lattice — ordered levels (Top Secret > Secret > Confidential, or High integrity > Low). Up and down only make sense on a lattice.

Simple vs Star (*) — Simple is always the read rule. Star is always the write rule. The outline’s “Star Model” is that write rule, not a third lattice.

Reference monitor — the abstract enforcer of every access. Its implementation is the security kernel. Well-known CBK: always invoked, tamperproof, small enough to verify.

Fail securely — on failure the system moves to a more-secure (usually closed) state. A crashed firewall that passes traffic failed open.

Fail-safe — on failure, protect human life (unlock the fire door). Do not swap these two on a life-safety stem.

Part 2 · 3.1 principles (official list, one-line each)

Objective 3.1 is not a vibe. It is eleven named ideas. A FIRST stem almost always names one of them.

Principle (3.1)If X, then YClassic miss
Threat modelingIf you have not enumerated abuse, you are guessingBuying a control before STRIDE / attack trees
Least privilegeIf the job needs read, do not grant adminShared “break-glass” that never expires
Defense in depthIf one layer dies, the next still holdsOne “silver bullet” product
Secure defaultsIf nobody configures it, it stays lockedShip open, harden later
Fail securelyIf the box crashes, deny (unless life-safety)Fail-open “for availability”
Segregation of duties (SoD)If one person can create and approve, fraud is one ticketSame engineer writes and releases
Keep it simple and smallIf the TCB is huge, you cannot verify itFeature sprawl as “flexibility”
Zero trust / trust but verifyIf location is the only check, lateral movement is free“Inside the firewall is trusted”
Privacy by designIf privacy is bolted on, it will leakCollect first, minimise later
Shared responsibilityIf the CSP patches the hypervisor, you still own the data and IAM“It is in the cloud, so they secure it”
SASEIf users are everywhere, security follows identity, not the castleTreating SASE as a LAN replacement
Common trap

Do not confuse fail-secure with fail-safe. Fail-secure prioritises confidentiality and integrity (lock the door). Fail-safe prioritises human life (unlock the door so people leave a fire). The exam tests which goal the stem privileges.

Part 3 · 3.2 models — pick the property, then the name

The official examples are Biba, Star Model, Bell-LaPadula. Well-known CBK that the CAT still uses: Clark-Wilson (commercial integrity) and Brewer-Nash / Chinese Wall (conflict of interest). Learn the goal, not a mnemonic tattoo.

Flow 1 · Simple = read, Star = write
Bell-LaPadula · confidentiality Biba · integrity Top Secret / High Secret / Mid · you are here Confidential / Low High integrity Mid integrity · you are here Low integrity Simple: no read up Star (*): no write down Simple: no read down Star (*): no write up Stops leaks of classified data Stops dirty data contaminating clean Strong star (both models): read and write only at your own level. Discretionary matrix can still deny a same-level access.

Read both columns. Simple never becomes a write rule. Reverse up/down when you switch from secrets to integrity.

ModelProtectsRule you must sayUse when
Bell-LaPadulaConfidentialityNo read up (Simple). No write down (Star).Military / classified labels
Star Model (*)The write ruleStar = write. BLP star blocks downward leaks. Biba star blocks upward contamination.Any stem that is about writing, not reading
BibaIntegrity (lattice)No read down. No write up. Invocation: do not call a higher-integrity subject.Do not let dirty data poison clean
Clark-WilsonCommercial integritySubject → transformation procedure → constrained data item. SoD + audit.Ledgers, payments, well-formed transactions
Brewer-NashConflict of interestAccess history dynamically walls you off from a competitor’s dataset.Consultants, banks, law firms
Say this out loud

Simple is read. Star is write. BLP stops leaks. Biba stops dirt. Clark-Wilson forces an approved program and two people. Brewer-Nash remembers who you already touched.

3. Decision flow — pick the property

Decision · the first diamond is the property, not the product
A glowing decision diamond splitting into Path A and Path B
Caption, not the art: you only reach “buy the WAF” after you know whether the stem is a leak, a poison, a conflict, or a safety-critical fail-open. Skipping the property is the engineer-brain miss.
Flow 2 · from ticket to model
New system, API, plant, or facility Name the asset + owner Life-safety on failure? yes Fail-safe unlock / escape no · security first What must not happen? Leak of a secret BLP + Star no write down Dirty data up Biba + Star no write up Fraud on a ledger Clark-Wilson + SoD Conflict of interest Brewer-Nash / Chinese Wall Then apply 3.1: secure defaults, least privilege, fail securely, keep it small.

Read top → bottom. Diamond = decision. Green border = commercial integrity or life-safety, not a lattice slogan. “Ignore” is not a model.

4. How to choose crypto and site controls

Objective 3.6 is select cryptographic solutions: lifecycle, methods (symmetric, asymmetric, elliptic curves, quantum), and PKI including quantum key distribution. Objective 3.7 is the attack list. Objectives 3.8–3.9 are the site. Do not mix the three jobs.

Feel · symmetric, hybrid, PQC, TPM
Pipeline from a shared key through a hybrid lock to a post-quantum vault and a TPM chip
Bulk data is fast and symmetric. The wrap that carries the session key is asymmetric or PQC. The chip that seals the key is a TPM or HSM. Exact names are in the tables.

Crypto menu (outline words, not vendor words)

JobPreferWhyDo not pick
Encrypt 500 GB tonightAES (FIPS 197), authenticated mode (GCM)Symmetric is fast; one shared keyRSA on the whole file
Carry that AES key over an untrusted pathRecipient’s public key, or ML-KEM (FIPS 203)Asymmetric / KEM solves distributionEmailing the AES key in the clear; sharing a private key
Prove who signed and they cannot deny itDigital signature (FIPS 186-5, or ML-DSA FIPS 204)Only a uniquely held private key gives non-repudiationEncryption alone; a shared signing key
Integrity of a file, no identity claimSHA-2 / SHA-3 hash, or HMAC if you share a keyOne-way fingerprintCalling a hash “encryption”
Long-lived data, harvest-now-decrypt-laterCrypto-agile path to FIPS 203 / 204 / 205Shor breaks RSA/ECC; Grover only halves symmetric strength“We use AES-256, so RSA-2048 is fine forever”
Physics-layer key agreement (outline names it)Quantum key distribution (QKD)Detect eavesdropping on the quantum channelTreating QKD as a drop-in for TLS certificates
NIST PQC — official names, August 2024

FIPS 203 ML-KEM (from CRYSTALS-Kyber) — key encapsulation; parameter sets ML-KEM-512 / 768 / 1024. FIPS 204 ML-DSA (from Dilithium) — module-lattice signatures. FIPS 205 SLH-DSA (from SPHINCS+) — stateless hash-based signatures, a non-lattice fallback. Source: NIST PQC FIPS approval. Hybrid (classical + PQC) is a migration pattern, not a replacement for key custody.

Cryptanalytic attacks (3.7 — official list)

AttackWhat the attacker has / doesFirst defence
Brute forceTries every keyAdequate key length; rate-limit
Ciphertext-onlyOnly intercepted ciphertextModern ciphers; no classic substitution
Known plaintextSome plaintext + matching ciphertextDo not reuse keys / IVs
Frequency analysisLetter / block statisticsNot a threat to AES; kills Caesar / Vigenère
Chosen ciphertextFeeds crafted ciphertext, watches decryptAuthenticated encryption; do not leak padding oracles
Implementation / side-channel / timing / faultPower, cache, clock, injected faultsConstant-time code; HSM; physical shielding
MITMSits between partiesAuthenticated key exchange; pin / validate certs
Pass the hash / Kerberos exploitationReplays a hash or ticket, never the passwordCredential guard, short ticket life, no NTLM
RansomwareAdversary-controlled encryption of your dataBackups offline; least privilege; restore test

Site and facility (3.8–3.9)

Apply the same layered idea to dirt and concrete. Concentric rings: perimeter, building, interior, data hall. CPTED (well-known CBK): natural surveillance, natural access control, territorial reinforcement. Functional goals: deter, detect, delay, respond.

Outline objectControlExam judgment
Wiring closet / IDFLocked, logged, no public corridor latchA closet is a network core, not a broom room
Server room / data centreMantrap, badge + PIN, cameras, raised floor / hot-aisleTwo-person integrity on the inner door if the stem is high-value
Media / evidence storageDual control, chain of custody, fire-rated media vaultEvidence is a legal object, not a backup tape pile
HVAC / environmentIndependent cooling, water sensors, ASHRAE-range temperatureHeat is an availability attack
PowerUPS then generator; redundant feedsUPS is ride-through. Generator is sustainment. Both, not either.
Fire — occupied + electronicsClean agent (e.g. FM-200 / HFC-227ea, Novec 1230 / FK-5-1-12)Water destroys gear. CO2 displaces oxygen and can kill staff.
Fire classes (CBK)A ordinary, B liquids, C electrical, D metals, K cooking oilsDo not put water on a live Class C rack
Exam trap

“Encryption makes us quantum-safe.” Symmetric AES-256 still has a huge work factor after Grover. RSA and ECC do not survive Shor. Long-lived records that were wrapped with RSA-2048 are the harvest-now-decrypt-later case. QKD is a named PKI-adjacent method in the outline; it is not the same as ML-KEM.

5. Runbook — manager, then engineer, then prove

This is not a CLI lab. It is the architecture-board sequence ISC2 wants when the stem says FIRST or BEST after a design failure. Source for the manager half: official 3.1 + 3.10 (lifecycle). Source for crypto: NIST SP 800-57 Part 1 and FIPS 203/204/205. Source for the site: official 3.8–3.9.

Side A — Manager / CISO (scope before spend)

  1. Name the asset, the owner, and the property

    Ledger integrity, classified design, conflicted client file, occupied hall — not “the network.” The data owner classifies. Security does not invent the label.

  2. Threat-model before the purchase

    STRIDE (spoof, tamper, repudiate, info-disclose, DoS, elevate) is the well-known whiteboard. If this sheet does not exist, any WAF or HSM purchase is decoration.

  3. Pick the principle and the model from residual risk

    SoD + Clark-Wilson if one person can commit fraud. BLP if a downward write leaks a secret. Fail-safe if a locked door kills people. Record the residual risk the owner accepts.

  4. Write the standard, then fund the tool

    Policy → standard → procedure. “Buy a next-gen WAF” is not a first move. “Payments change only through an approved TP, two people, fail closed” is.

Side B — Engineer (design the control)

  1. Map the architecture element (3.5)

    Client, server, database, crypto module, ICS, cloud (IaaS/PaaS/SaaS — shared responsibility), container, serverless, IoT, embedded, HPC, edge, virtualized. The vuln class follows the element: VM escape is a hypervisor problem; a poisoned training set is a supply-chain problem; an exposed PLC is an ICS problem.

  2. Implement IS capabilities (3.4)

    Memory protection: rings, ASLR, NX/DEP, process isolation. TPM (ISO/IEC 11889): hardware root of trust, measured boot, sealed keys, attestation. Encryption/decryption: keys in an HSM or TPM, never in a shared folder. FIPS 140-3 is the module validation program — do not invent a “Level 5.”

  3. Build the crypto and the room you were given

    AES-GCM for bulk. Public-key or ML-KEM wrap for the session key. Signatures with a uniquely held private key. Occupied data hall: clean agent + UPS + generator. Closets locked. Evidence under dual control.

  4. Close the TOCTOU and covert-channel windows

    Check-and-use must be one locked operation. Covert storage and timing channels are unintended paths — rate-limit, isolate shared resources, audit unusual clock or temp-file use.

Side C — Prove (pilot, then expand)

  1. Pick one fail-closed story

    A single engineer cannot both create and approve a vendor. A Secret subject cannot write a Top Secret object into a Confidential share. A crashed policy engine denies. Write those as pass/fail before go-live.

  2. Instrument the path

    Access-triple logs (who, which TP, which CDI). TPM attestation on boot. Key-lifecycle record (generate, distribute, store, use, rotate, archive, revoke, destroy — SP 800-57). Clean-agent system test and UPS transfer test.

  3. Break it on purpose

    Attempt the forbidden write, the shared signing key, the fail-open reboot. Expect deny + log. If it succeeds, the model was a costume.

Pilot evidence — fields a reviewer should see
asset=vendor-ledger
owner=finance-controller
property=integrity+SoD
model=clark-wilson
tp=create-vendor | approve-payment
subject_create != subject_approve
action=deny
reason=same-subject-both-roles
fail_mode=closed
aes_at_rest=AES-256-GCM
key_wrap=ML-KEM-768 | RSA-OAEP (migration flagged)
sig=ML-DSA | FIPS-186-5
privkey_custody=hsm-partition unique
tpm=measured-boot attest=pass
fire=clean-agent occupied=yes
ups=transfer-ok generator=tested

6. Runtime path — keys, TPM, facility

Two pictures after go-live. First the cryptographic object. Then the building that keeps the module honest. Objective 3.10 is the lifecycle that wraps both: stakeholder needs → requirements → architectural design → development → integration → verification and validation → transition → operations / sustainment → retirement / disposal.

Proof · the desk and the aisle that close the ticket
Operations desk and data-center aisle with health checks confirming the pilot path
Green is not a vibe. Green is deny on the forbidden role pair, attestation pass, key-destruction record, clean-agent and UPS tests dated.
Flow 3 · key object and concentric site
Generate Distribute Use Rotate Revoke Destroy SP 800-57 key lifecycle · store and archive sit under Use · never skip Destroy Perimeter · fence, lighting, CPTED Building · badge, mantrap, cameras Interior · IDF locks, restricted work area Data hall HSM · TPM · clean agent · UPS A stolen laptop outside the inner ring must not hold the unsealed private key.

Keys live a lifecycle. The building is a set of rings. Retirement (3.10) includes crypto-shred and media destruction, not just unplugging a rack.

Shared responsibility (3.1)

IaaS: you own OS, data, IAM, crypto. PaaS: you own the app and data. SaaS: you own identity, classification, and what you put in. The CSP’s SOC 2 does not accept your residual risk.

SASE (3.1)

Secure access service edge is identity-following access, not a new castle. It does not write the Clark-Wilson triple inside the ledger, and it does not pick your fire agent.

7. Traps and proof

TrapWhat people sayWhat is actually trueFix
Arrows without a property“No read up, no write down” for every stemThat is BLP confidentiality. Integrity flips it.Ask what must not happen, then name the model.
Biba as a ledger“No write up covers payments.”Biba is a lattice. It does not force a well-formed TP or SoD.Clark-Wilson access triple + audit.
Fail-open as availability“If the firewall dies, pass traffic.”That is fail-open. 3.1 says fail securely unless life-safety.Fail-closed for secrets; fail-safe for people.
RSA on bulk data“Stronger key, encrypt the warehouse.”Asymmetric is for wrap and sign. Symmetric is for bulk.Hybrid: AES + public-key / ML-KEM wrap.
Shared signing key“The team mailbox signs contracts.”Non-repudiation dies the moment two people can sign.Unique private key in an HSM partition.
Quantum-safe slogan“We encrypt, so we are fine.”Shor kills RSA/ECC. Harvest-now is a 3.6 problem today.Inventory + crypto-agility + FIPS 203/204/205.
Water or CO2 in an occupied hall“Sprinklers are code; CO2 is clean.”Water kills electronics. CO2 can kill people.Clean agent for occupied + electronics.
Cloud = their problem“Shared responsibility means they patch everything.”You still own data, keys, IAM, and classification.Write the split. Accept residual risk in writing.
Pilot checklist — print this
Interview pair — weak vs strong

Weak: “Domain 3 is Bell-LaPadula and AES.” Strong: “I name the property first. Secrets get BLP — simple no-read-up, star no-write-down. Ledgers get Clark-Wilson and two people. Bulk data is AES; the wrap is asymmetric or ML-KEM; the signature key is uniquely held. The room fails safe for people and the control plane fails closed for data. I can show the deny log, the attestation, and the key-destruction record.”

Knowledge check

Six judgment items. Same traps the CAT uses: fail-secure vs fail-safe, Star as write, hybrid crypto, Biba as a fake ledger, occupied-room fire, harvest-now-decrypt-later. Check answers, then reset if you miss any.

Q1

A policy engine crashes during a change window. Leadership wants “availability first.” Which 3.1 behaviour is correct for a confidentiality-critical control plane that is not a life-safety door?

Correct: a. Fail securely means a failure drives the system to a more-secure state. Fail-safe is the life-safety exception (unlock so people escape). Least privilege and SASE do not answer a crash. Re-read Mental model.
Q2

A defence contractor needs a model that stops a Secret-cleared engineer from copying a Top Secret design into a Confidential shared folder. Which rule must the system enforce?

Correct: c. The leak is writing sensitive data down to a lower classification — BLP Star (no write down). Simple blocks reading, not this write. Biba and Clark-Wilson are integrity models. Re-read models and Decision flow.
Q3

A fintech team must encrypt 500 GB of nightly database backups quickly, and also carry the key across an untrusted link. Which design matches CISSP practice?

Correct: b. Hybrid: fast symmetric for bulk, asymmetric or ML-KEM for distribution. RSA on 500 GB is the wrong job. A hash is not encryption. Private keys are never shared. Re-read How to choose.
Q4

Auditors want users to change ledger records only through approved programs, with separation of duties and full logging — never by editing files. A new hire says Biba “no write up” already covers this. Why is Biba alone insufficient?

Correct: b. Biba is a lattice. Clark-Wilson adds the subject–TP–CDI triple, SoD, and audit. Biba is an integrity model, not a confidentiality model. Re-read the model table and Decision flow.
Q5

A new data-centre room will have engineers working inside the racks. You must pick fire suppression that protects both staff and servers. Which choice fits?

Correct: d. Clean agents suppress without drowning electronics and are used in occupied spaces. Water destroys equipment. CO2 can asphyxiate staff. Manual-only is not a suppression system. Re-read site and facility.
Q6

Twenty-five-year loan records travel under RSA-2048 TLS and sit encrypted at rest. Management says “encryption makes them quantum-safe.” Which risk is most under-addressed?

Correct: b. Shor breaks RSA/ECC. Twenty-five-year data outlives that horizon. AES-256 only loses roughly half its bits to Grover and is not “trivial.” TLS 1.3 is the living standard. A TPM seals runtime keys; it does not un-harvest last year’s ciphertext. Re-read crypto menu and runtime path.

Sources

Related: Domain 2 · Asset Security · Domain 4 · Communication and Network Security · All 8 domains map · Domain 3 timed assessment · 8-week roadmap