Insecure design: a flaw that a request filter can only slow down

Insecure design is the weakness that remains when the code does exactly what it was designed to do and the design was wrong: a control was never specified, or was specified badly. It is A06 in the OWASP Top 10:2025. A WAF can slow abuse of such a flaw and can hold a known request shape back for a while, but it cannot give the application the control it lacks.

Section Rules Updated OWASP A06:2025

Quick facts

Category
A06:2025 Insecure Design, down from fourth place in 2021
Mapped weaknesses
39 CWEs, among them CWE-602 client-side enforcement of server-side security, CWE-799 improper control of interaction frequency and CWE-841 improper enforcement of behavioral workflow
Occurrences in OWASP's data
729,882 occurrences and 7,647 CVEs
Incidence rate
Up to 22.18% of tested applications, 1.86% on average across the mapped CWEs
Main remedy
Threat modeling, security requirements in user stories, vetted design patterns, tests for misuse cases
Does a WAF stop it?
Not the flaw. Rate limits can slow abuse and a rule can act as a temporary virtual patch; the design has to change

Figures are from the A06:2025 page of the OWASP Top 10.

Design flaw or implementation flaw

OWASP describes the category as missing or ineffective control design, and it is explicit that insecure design is not the same thing as insecure implementation. The two have different root causes, appear at different stages and need different fixes. A flawless implementation cannot repair a design that never included the control needed to stop a particular attack.

The difference shows in what a fix looks like. A SQL injection is a correct design implemented carelessly: the specification said to look up an account, and the code joined strings. Parameterising the query closes it. A password reset that asks for a mother's maiden name is implemented exactly as specified, and the specification is the weakness. OWASP names that case itself, and it notes that NIST SP 800-63B and the OWASP ASVS discourage security questions because several people may know the answers.

OWASP also points at a root cause that sits before design: not profiling the business risk of the system, and so never deciding how much security design it needs. The reverse of that decision gives the vocabulary of this category, which is requirements first, then design, then development.

Threat modeling

The OWASP Threat Modeling Cheat Sheet defines threat modeling as a structured, repeatable process for gaining actionable insights into the security characteristics of a system. It reduces to four questions taken from the Threat Modeling Manifesto: what are we working on, what can go wrong, what are we going to do about it, and did we do a good enough job.

A06 asks for it as a continuing practice, not a workshop held once. Its guidance is specific about where to spend it: high-risk areas such as authentication, access control, business logic and key workflows. Define the expected flow and the failure flows, check the assumptions, and write the result into user stories so that a control becomes a requirement someone can test.

The output that matters for operations is small. A list of the abuse cases for each critical flow says which requests are worth watching. If the model says that a booking endpoint can be hammered to reserve inventory, that endpoint is where a limit belongs, and it is the first place to look in the activity log.

Secure design patterns

OWASP's prevention list for A06 is mostly about process, with a few concrete controls.

  • A secure development lifecycle. Application security specialists take part from the start of a project through maintenance, to help design security and privacy controls.
  • A library of vetted patterns. Paved-road components mean a team reuses an authentication or authorisation design that was already reviewed.
  • Plausibility checks at every tier. The back end validates what the front end claims, so a price, role or step count sent by the browser is never trusted on its own.
  • Tests for misuse. Unit and integration tests confirm that critical flows resist the threat model, covering misuse cases as well as intended use.
  • Separation by exposure. System and network tiers are split according to exposure and protection needs, and tenant isolation is built into every tier.

Several of these are the same idea as zero trust applied to application logic: no tier accepts a claim because of where it came from.

The CWEs under A06

The 39 weaknesses mapped to A06 are a mixed set, which is what you would expect from a category defined by cause and not by symptom. A few groups stand out.

  • Trusting the client. CWE-602 client-side enforcement of server-side security, CWE-472 external control of an assumed-immutable web parameter, CWE-807 reliance on untrusted inputs in a security decision and CWE-642 external control of critical state data.
  • Flows that can be reordered or repeated. CWE-841 improper enforcement of behavioral workflow and CWE-362 race condition. These are the territory of business logic flaws.
  • No limit on use. CWE-799 improper control of interaction frequency, which MITRE defines as the product not properly limiting the number or frequency of its interactions with an actor, such as the number of incoming requests.
  • Credentials and sensitive data. CWE-256, CWE-522 and the cleartext storage entries CWE-311, CWE-312 and CWE-313. The storage side overlaps cryptographic failures.
  • Protocol and upload design. CWE-444 inconsistent interpretation of HTTP requests, covered in request smuggling, CWE-434 unrestricted upload of a file with a dangerous type, covered in file upload attacks, and CWE-1021 improper restriction of rendered UI layers, covered in clickjacking.
  • Structure. CWE-657 violation of secure design principles, CWE-653 insufficient compartmentalization, CWE-1125 excessive attack surface and CWE-693 protection mechanism failure.

The full list is on the OWASP page; the MITRE entry for CWE-799 is the one most relevant to a rate limit.

Three cases from OWASP

OWASP gives three example scenarios. Each one is worth reading with a single question in mind: what would a filter in front of the application see?

Security questions for password recovery

The recovery flow relies on questions. Nothing malformed arrives, and the answers are correct. OWASP's remedy is to remove the feature and replace it with a stronger design. A filter has nothing to match, and a limit on attempts does not change the fact that the answers can be known.

Group booking at a cinema chain

The chain offers group discounts and asks for a deposit only above fifteen attendees. An attacker could book a very large number of seats across cinemas in a few requests, with major revenue loss. This one a limit can touch, because the abuse depends on speed and repetition. It does not repair the missing business rule, such as a cap on total held seats, and a patient attacker stays under any threshold.

Bots buying scarce stock

A retailer with no defence against bots that buy high-demand graphics cards for resale suffers in reputation and with customers. OWASP suggests anti-bot controls and business rules, such as flagging purchases made seconds after stock becomes available. Limits help here and are part of the answer; see scalpers for the pattern. The rule about purchase timing is a decision for the application.

Where a WAF helps and where it does not

What it can do

  • Slow repeated use of a flow. A rate limit counts requests against a subject and refuses them past a threshold. That addresses the interaction-frequency weakness directly, and it blunts mass booking, enumeration and automated purchasing. Who is counted matters as much as the number; see limits by address, user or key.
  • Hold a known request shape back as a virtual patch. If a design flaw has a visible shape, such as an endpoint or parameter the application should never have exposed, a rule can refuse it while the redesign is under way. It is temporary by definition: it covers the shape somebody anticipated and nothing around it.
  • Show the traffic. The log records which routes receive bursts and which requests a rule would refuse, which turns a threat model into numbers.

What it cannot do

  • Supply a missing control. A deposit rule, a workflow check or a server-side price lookup lives in the application. A filter does not know your business rules.
  • See the flaw in the request. The requests are well formed and often correct. This is why business logic flaws are the hardest to catch from outside.
  • Outlast a slow attacker. A limit refuses requests over a threshold. Requests under it, spread over time or across subjects, are still served.
  • Tell abuse from use. A limit refuses valid requests because there were many of them, and a busy legitimate customer can be caught by the same count.

Order of work

Start in monitor mode. Put the limit on the flow the threat model flagged, read the log for a few days of ordinary traffic, find who would have been refused, and only then move that rule to block through a staged rollout. In parallel, file the design change. Treat any rule you added as a note that the work is not finished and remove it when the flow is redesigned.

The neighbouring categories show the same boundary from other sides: broken access control is a missing server-side check, and security misconfiguration is a setting that was left wrong where a design flaw is a control that was never there.

Questions

What is insecure design in the OWASP Top 10?

It is A06:2025: weaknesses where a security control was missing or designed ineffectively. OWASP separates it from insecure implementation, because a perfect implementation of a flawed design still leaves the flaw.

Is insecure design the same as a coding bug?

No. A coding bug is a correct design implemented wrongly, and a fix in the code closes it. An insecure design does what it was designed to do, so the design has to change, and no amount of careful coding repairs it.

Can a WAF fix insecure design?

No. A WAF can slow abuse of a flawed flow with rate limits and can refuse a known request shape as a temporary virtual patch. The flaw stays in the application until its design is changed.

What is threat modeling and why does OWASP recommend it?

Threat modeling is a structured, repeatable way to examine a system from an attacker's point of view. OWASP recommends it for high-risk areas such as authentication, access control, business logic and key workflows, because it finds missing controls before they are built.

Which WAF mode should I use first for a design flaw?

Monitor. It records matches and serves every request, so you can see which routes receive the traffic a limit or rule would refuse before block mode refuses any of it.