Security Models & Secure Architecture
CISSP CISSP · Security Architecture and Engineering
Why models matter
Security models are formal descriptions of how a system should enforce security policy. They matter for the CISSP because they turn vague ideas like "keep secrets secret" into precise rules that a system can implement and an examiner can test. The exam expects you to know the major models, what property each protects (confidentiality, integrity, or availability), and the rules each one enforces. The two you must know cold are Bell-LaPadula (confidentiality) and Biba (integrity), because the exam tests them in scenarios where you must apply their rules, not just name them.
Bell-LaPadula: confidentiality
The Bell-LaPadula model is a state machine model designed for confidentiality, originally developed for the US military. It has two core properties. The Simple Security Property — "no read up" — states that a subject cannot read an object at a higher security level than the subject's clearance. The *-Property (star property) — "no write down" — states that a subject cannot write to an object at a lower security level than the subject's clearance, because doing so could leak information downward. Together, these rules ensure information flows only upward or at the same level: a Secret-cleared user can read Secret and lower documents, and can write to Secret and higher documents, but cannot read Top Secret documents or write to Unclassified ones. The model also has a strong star property variant (read and write only at the same level). Bell-LaPadula addresses confidentiality only — it says nothing about integrity, and it does not address covert channels.
Biba: integrity
The Biba model is the mirror image of Bell-LaPadula, designed for integrity. Its rules are the reverse: the Simple Integrity Property — "no read down" — states that a subject cannot read an object at a lower integrity level, because low-integrity data could corrupt the subject's work. The *-Integrity Property — "no write up" — states that a subject cannot write to an object at a higher integrity level, because a lower-integrity subject could corrupt higher-integrity data. The exam's favourite trick is to swap the models: a question about preventing the corruption of critical data is a Biba question, not a Bell-LaPadula question. Remember it as: Bell-LaPadula protects secrets (confidentiality), Biba protects trustworthiness (integrity).
Other models
Clark-Wilson is an integrity model for commercial environments, built around well-formed transactions and separation of duties: data can only be modified through authorised programs (transformation procedures), and the person who certifies a transaction cannot be the person who executes it. It introduces the concepts of constrained data items (data protected by the model) and unconstrained data items (everything else). The Brewer and Nash model (the Chinese Wall) addresses conflicts of interest: once a user accesses data from one company in a conflict set, they cannot access data from a competing company in the same set — the classic example is a consultancy that serves rival clients. The Graham-Denning model describes how subjects and objects are created and deleted and how access rights are transferred. Take-Grant models the transfer of rights between subjects. The exam tests Clark-Wilson and Brewer-Nash at a recognition level and Bell-LaPadula and Biba at an application level.
| Model | Protects | Key rule |
|---|---|---|
| Bell-LaPadula | Confidentiality | No read up, no write down |
| Biba | Integrity | No read down, no write up |
| Clark-Wilson | Integrity (commercial) | Well-formed transactions, separation of duties |
| Brewer and Nash (Chinese Wall) | Confidentiality (conflicts of interest) | No access to competing datasets in the same conflict set |
| Graham-Denning | Access control | Rules for creating/deleting subjects and objects, transferring rights |
Secure design principles
Domain 3 also covers the principles that guide secure system design, drawn largely from the classic Saltzer and Schroeder paper. Least privilege means every subject gets only the access it needs. Defence in depth means layering controls so that the failure of one does not compromise the system. Fail-secure (fail closed) means a system that fails does so in a secure state — a firewall that fails should block traffic, not pass it. Separation of duties and separation of privilege divide sensitive operations so no single person or condition can complete them alone. Complete mediation means every access is checked every time, with no caching of authorisation decisions. Open design means the security of a system must not depend on the secrecy of its design — Kerckhoffs's principle: only the key is secret, never the algorithm. Least common mechanism warns against sharing mechanisms between users, since shared mechanisms create shared vulnerabilities. Psychological acceptability means security controls must be usable enough that people actually use them rather than working around them. The exam tests these as scenario judgements: a design that fails open, or that depends on keeping its algorithm secret, violates a named principle.
Trusted computing and the reference monitor
The reference monitor is an abstract concept: a component that mediates all access between subjects and objects, is tamper-proof, is always invoked, and is small enough to be verified. The security kernel is the implementation of the reference monitor in an operating system, and the Trusted Computing Base (TCB) is the totality of the hardware, firmware and software that enforces the security policy — everything inside the TCB must be trusted, and everything outside it is untrusted. A smaller TCB is better, because there is less to verify and less that can go wrong. Related concepts include security rings (privilege levels in a CPU, with Ring 0 the most privileged — the kernel — and Ring 3 the least, where user applications run), process isolation (each process runs in its own memory space), and trusted platform modules (TPMs), hardware chips that provide secure key storage, attestation, and measured boot.
Vulnerabilities in system design
The exam expects familiarity with the vulnerability classes that arise from poor design. Covert channels are communication paths not intended for information transfer — a covert timing channel encodes data in the timing of events, and a covert storage channel encodes data in shared storage. TOCTOU (time-of-check to time-of-use) is a race condition where the state of a resource changes between the security check and its use. Buffer overflows occur when a program writes more data to a buffer than it can hold, overwriting adjacent memory — the classic exploitation primitive. Side-channel attacks extract information from physical characteristics such as power consumption, electromagnetic emissions, or timing (the Spectre and Meltdown vulnerabilities are examples of timing side channels in modern processors). Maintenance hooks and backdoors are undocumented access paths left in systems. The exam tests these at a level where you must recognise the vulnerability from a description and know the appropriate countermeasure.
Evaluation criteria
Security evaluation frameworks define how systems are assessed. The Common Criteria (ISO/IEC 15408) is the current international standard, using Evaluation Assurance Levels (EALs) from EAL1 (functionally tested) to EAL7 (formally verified design and tested). A Protection Profile (PP) defines the security requirements for a category of product, and a Security Target (ST) is a vendor's claim about a specific product's security. The older TCSEC (the Orange Book) defined classes from D (minimal protection) through C1, C2, B1, B2, B3 to A1 (verified design), and ITSEC was the European predecessor to the Common Criteria. The exam tests Common Criteria concepts most, with TCSEC appearing mainly as historical context and in questions about its class structure.
Exam mindset for security models
Model questions are among the most predictable on the exam: learn the rules of Bell-LaPadula and Biba well enough to apply them to a scenario, know what each other model protects, and be ready to name the design principle a scenario violates. If a question describes information flowing downward to a less-cleared user, think Bell-LaPadula; if it describes low-quality data corrupting high-quality data, think Biba.
Exam traps to watch for
- Swapping Bell-LaPadula and Biba: Bell-LaPadula is confidentiality (no read up, no write down); Biba is integrity (no read down, no write up).
- Thinking Bell-LaPadula addresses integrity or covert channels — it addresses confidentiality only.
- Confusing fail-secure with fail-safe: fail-secure (fail closed) blocks access on failure; a firewall that passes traffic when it fails has failed open, which is insecure.
- Assuming security through obscurity is acceptable: open design (Kerckhoffs's principle) requires that security rest on the key, not on keeping the design secret.
Practice questions
Test yourself on this guide. Answers are checked instantly, with a full explanation for each question.
Q1.An analyst with Secret clearance attempts to read a Top Secret document and to write a report to an Unclassified shared folder. Under the Bell-LaPadula model, which actions are permitted?
Explanation: Bell-LaPadula forbids both: "no read up" prevents a Secret-cleared subject from reading Top Secret data, and "no write down" prevents writing to a lower-classification object, which could leak Secret information downward. The analyst may read at Secret and below and write at Secret and above. Both options that permit one of the actions violate one of the two properties.
Exam objective: CISSP Domain 3: Understand security models (Bell-LaPadula)
Q2.A financial system must ensure that data entered by junior clerks cannot corrupt the verified transaction ledger used by auditors. Which security model BEST addresses this requirement?
Explanation: This is an integrity problem: low-integrity input must not corrupt high-integrity data, and Biba's *-integrity property ("no write up") directly prevents a lower-integrity subject from writing to a higher-integrity object. Bell-LaPadula addresses confidentiality (reading secrets), not integrity. Brewer and Nash addresses conflicts of interest between competing datasets. Take-Grant models the transfer of rights, not integrity levels.
Exam objective: CISSP Domain 3: Understand security models (Biba)
Q3.A consulting firm serves two competing pharmaceutical companies. A consultant who has worked on Company A’s account requests access to Company B’s project files. Which model governs this situation?
Explanation: The Brewer and Nash (Chinese Wall) model exists for exactly this conflict-of-interest scenario: once a subject accesses data from one organisation in a conflict set, access to competing organisations in the same set is denied. Clark-Wilson concerns well-formed transactions and separation of duties in commercial integrity, not conflicts of interest. Bell-LaPadula concerns classification levels, which are not the issue here. Biba concerns integrity levels, not competing clients.
Exam objective: CISSP Domain 3: Understand security models (Brewer and Nash)
Q4.A firewall is configured so that if its inspection engine crashes, all network traffic is blocked until the engine restarts. Which secure design principle does this configuration follow?
Explanation: Fail-secure (fail closed) means a system that fails does so in a secure state — here, blocking traffic rather than passing it uninspected. Least privilege concerns the scope of access granted, not failure behaviour. Defence in depth is about layering controls, which is not what the scenario describes. Complete mediation means every access is checked every time; it does not describe what happens when the checking mechanism itself fails.
Exam objective: CISSP Domain 3: Understand secure design principles
Q5.A vendor claims its encryption product is secure because its algorithm is proprietary and has never been published. A security architect evaluating the product should be MOST concerned that this violates which principle?
Explanation: Open design (Kerckhoffs's principle) states that a system's security must not depend on keeping its design secret — only the key is secret. A proprietary, unpublished algorithm cannot be independently reviewed and is likely to contain weaknesses; "security through obscurity" is not security. Least common mechanism concerns shared mechanisms creating shared vulnerabilities, not algorithm secrecy. Psychological acceptability concerns usability. Separation of duties concerns dividing sensitive operations between people.
Exam objective: CISSP Domain 3: Understand secure design principles (open design)
Q6.Which component of a system is responsible for mediating ALL access between subjects and objects, and must be tamper-proof, always invoked, and small enough to be verified?
Explanation: The reference monitor is the abstract concept of a component that mediates all access and satisfies the three properties (tamper-proof, always invoked, verifiable); the security kernel is its implementation in an operating system. The TCB is the totality of hardware, firmware and software that enforces policy — a broader concept that includes the security kernel but is not itself the mediating component. A TPM is a hardware chip for key storage and attestation. An ACL is a data structure listing permissions, not a mediating component.
Exam objective: CISSP Domain 3: Understand trusted computing concepts (reference monitor, security kernel, TCB)
Q7.An attacker exploits a program by sending input longer than the allocated buffer, overwriting adjacent memory to execute malicious code. Which vulnerability class is this, and what is the MOST effective design-level countermeasure?
Explanation: Writing beyond a buffer’s allocated size is a buffer overflow, and the design-level fix is bounds checking (validating input length before copying) plus memory-safe coding practices. A covert timing channel encodes data in event timing, which is not described here. A TOCTOU race condition involves state changing between a check and its use. A side-channel attack extracts information from physical characteristics such as power or timing, not from memory corruption.
Exam objective: CISSP Domain 3: Understand vulnerabilities in system design (buffer overflow)
Q8.A government agency is procuring a firewall and requires it to be evaluated under the Common Criteria. The vendor provides a document describing the security claims for its specific product. What is this document called?
Explanation: A Security Target (ST) is the vendor's document stating the security properties claimed for a specific product (the Target of Evaluation). A Protection Profile (PP) defines requirements for a product category, not a specific product. An EAL is a rating level (EAL1–EAL7) describing the rigour of the evaluation, not a document. A certification report is produced by the evaluation scheme after evaluation, not by the vendor as its claim.
Exam objective: CISSP Domain 3: Understand security evaluation criteria (Common Criteria)