Skip to content
Tutorials Center
My learning

Topics

Lesson 25 of 80 ยท Authentication and Access Control

Classic Security Models

Learn the classic formal security models: Bell-LaPadula for confidentiality, Biba for integrity, Clark-Wilson for transactions and Brewer-Nash for conflicts of interest.

An access-control policy says who may do what. A security model is a precise, often mathematical, description of a policy, written so that you can reason about it and prove it enforces a goal such as "secrets never flow to people without clearance". The models in this lesson were developed for military and financial systems in the 1970s to 1990s, and their ideas still shape modern systems: SELinux labels, database row-level security, separation of duties in banking software, and cloud tenant isolation.

Authorization and Access Control Models introduced DAC, MAC, RBAC and ABAC, and mentioned Bell-LaPadula and Biba in a few lines. Here you will study the classic models properly, see which goal each one protects, and implement their rules in Python so the logic becomes concrete.

Vocabulary

  • Subject: an active entity, such as a user or process, that requests access.
  • Object: a passive resource, such as a file, record or database table.
  • Security label: a classification on an object (for example Public, Internal, Confidential, Secret) and a clearance for a subject.
  • Policy: the rules governing which subject-object operations are allowed.
  • Reference monitor: the always-invoked, tamper-proof, verifiable component that mediates every access.

Each model targets one of the three goals from the CIA triad: Bell-LaPadula protects confidentiality, Biba and Clark-Wilson protect integrity, and Brewer-Nash prevents conflicts of interest.

Bell-LaPadula: confidentiality

Developed in the early 1970s for military data, Bell-LaPadula (BLP) stops information from flowing upward to people who should not see it and downward into less-protected places. Levels are ordered, for example Public < Internal < Confidential < Secret. It has two core rules:

  • Simple security property ("no read up"): a subject at level L may not read objects above L.
  • *-property ("no write down"): a subject at level L may not write to objects below L. Otherwise a Secret-cleared user, or malware running as one, could copy secrets into a Public file.

A third rule, the discretionary security property, lets an access matrix further restrict access within what the labels allow. A weakness: BLP says nothing about integrity, so a low-level user may "write up" and corrupt a high-level file. The tranquility principle (labels do not change while in use) keeps the rules consistent.

Biba: integrity

Biba is the mirror image, designed to keep trusted data from being contaminated by less-trusted data. Levels now describe integrity (how much we trust the data or program):

  • Simple integrity property ("no read down"): a subject may not read objects of lower integrity, so it is not corrupted by untrusted input.
  • *-integrity property ("no write up"): a subject may not write to objects of higher integrity.

Think of a web server process: input from the internet is low-integrity data, so it should not be able to overwrite the system's configuration. Modern parallels are mandatory integrity levels in Windows and sandboxing browsers.

Because the rules are opposite, BLP and Biba cannot both be enforced on the same labels at the same time without conflict. Real systems apply the one that matches the goal.

Seeing the rules in code

Python Bell-LaPadula and Biba checkers

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Every row gets opposite answers from the two models, which is the point: one stops leaks, the other stops corruption.

Clark-Wilson: integrity for business transactions

Biba blocks bad data flows, but it cannot express business rules like "a payment must be approved by someone other than the person who created it". Clark-Wilson (1987) models how real organizations protect integrity. It introduces:

  • Constrained data items (CDIs): data whose integrity matters, such as account balances.
  • Unconstrained data items (UDIs): raw input that has not been validated.
  • Transformation procedures (TPs): the only programs allowed to change CDIs, each certified to keep data consistent.
  • Integrity verification procedures (IVPs): checks that CDIs are in a valid state, such as "the sum of all accounts equals the ledger total".
  • Access triples: (user, TP, CDI) relationships that say which user may run which TP on which data. Users never touch CDIs directly; they go through TPs.

The model also enforces separation of duties: the person who initiates a transaction cannot be the one who approves it, and audit logging records every use of a TP.

This is exactly how well-built business software works: you call "transfer money" (a TP), not "update balance" (a raw write), and a transfer validates input, keeps both sides balanced and logs itself (Secure Architecture and Design Patterns).

Python Clark-Wilson: triples and separation of duties

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The balances changed only through certified procedures, the invariant held, and every attempt, including the denied ones, left an audit trail.

Brewer-Nash (Chinese Wall): conflicts of interest

A consulting firm may advise two competing banks, but an employee who has seen Bank A's data must not later see Bank B's. Brewer-Nash is a dynamic model: the access you are allowed depends on your history. Data is grouped into company datasets, and datasets of competing companies are in the same conflict-of-interest class. Once you access one dataset in a class, you are barred from the others in that class (but you may access any company in other classes). The model appears in consulting, legal, audit and financial-advice settings, and in multi-tenant systems where a user must not mix tenants.

Other models worth knowing

Model Goal Core idea
Access control matrix (Lampson) Describe all permissions A table of subjects x objects holding rights; ACLs are its columns, capabilities its rows
Take-Grant Analyze how rights spread A graph where subjects can take or grant rights; used to decide whether a right can ever leak
Graham-Denning Safe creation and deletion Eight primitive operations to create/delete subjects and objects and transfer rights
Lattice-based Generalize labels Labels form a lattice; information may only flow upward ("can flow to" ordering)
Noninterference No covert influence Actions of high-level users must not be observable by low-level users
Information flow Track data paths Prevents indirect leaks, such as through shared files or timing

Choosing a model

Choose by the goal, not by fashion:

  • Protecting secrecy of classified information: Bell-LaPadula-style labels.
  • Protecting integrity of system files or trusted components: Biba-style integrity levels.
  • Protecting business transactions (finance, inventory, medical records): Clark-Wilson with separation of duties and audit.
  • Preventing conflicts of interest in shared services: Brewer-Nash.
  • Everyday applications: RBAC or ABAC, informed by these ideas (Authorization and Access Control Models).

No model solves everything. Covert channels (leaks through timing, storage or resource use) and human error sit outside the formal rules, which is why you also need threat modeling and monitoring.

Check your understanding

Quick check ยท 4 questions

of 4 answered You scored / 4 ยท perfect!

  1. 1. A Secret-cleared user tries to save a report into a Public folder. Which Bell-LaPadula rule blocks it?

    The *-property forbids writing to a lower level, so secrets cannot leak into less-protected storage.
  2. 2. Which model is mainly about preventing low-trust data from corrupting high-trust data?

    Biba protects integrity with no read down and no write up.
  3. 3. In Clark-Wilson, how may a user change a constrained data item?

    Users never touch CDIs directly. Certified procedures keep the data consistent, and access triples control who may run each one.
  4. 4. What makes Brewer-Nash different from the other models?

    Once you have accessed one company's data in a conflict class, you are barred from its competitors, so the rules change with your past actions.

Summary

  • A security model is a precise statement of a policy that lets you reason about a goal.
  • Bell-LaPadula protects confidentiality (no read up, no write down); Biba protects integrity (no read down, no write up); they conflict if applied to the same labels.
  • Clark-Wilson protects integrity in business systems with constrained data, certified transformation procedures, access triples, separation of duties and audit.
  • Brewer-Nash (Chinese Wall) uses access history to prevent conflicts of interest.
  • Other models (access matrix, Take-Grant, lattice, noninterference) analyze permission spread and information flow.
  • Pick the model that matches the goal, and remember covert channels and people lie outside the formal rules.

Next: Kerberos, LDAP and RADIUS, to see how users and services prove identity inside an organization.