Table of contents
Title
Table of content
Table of contents
Title

Dynamic Application Security Testing (DAST)

Updated

Dynamic application security testing (DAST) looks for security vulnerabilities in a running application. The scanner needs no source code: it sends requests to live endpoints and reads what comes back, the way an attacker probing from the outside would. OWASP notes that dynamic testing can be run by a tool or by a person, and that the two find different things.

Fluid Attacks’ DAST scanner tests seven attack surfaces: DNS records, headers, HTML content, SSL connections, unauthenticated HTTP endpoints, webhooks and WebSockets. It does not test authenticated endpoints. In Fluid Attacks’ State of Attacks 2026, built on 1,119,396 vulnerabilities reported across client systems during 2025, the most frequent weakness in applications was authentication mechanism absence or evasion, at a mean risk exposure of 855.99 CVSSF. The distance between what a scanner reaches and where the risk sits is the argument for everything after the scan.

This page covers what a DAST scan can reach and what it cannot, how its accuracy is measured, how Fluid Attacks runs it, how to wire it into a pipeline, and the point where manual testing takes over.

What is DAST?

DAST is a black-box security testing technique: it assesses an application while that application is running, with no view of the source code behind it. Because it works from the outside, it is language-independent and runs the same way against code you wrote and code you bought.

A scan has four stages: it maps and crawls the target to find pages, parameters and endpoints; sends attack payloads to the inputs it found; reads the responses for errors, unexpected content, timing changes or unauthorized access; and reports what it confirmed, with request and response as evidence.

What DAST is not

It is not a code reader. A DAST finding tells you an application behaves insecurely, not which line produced the behavior. That limitation is structural, not a product gap, and it has consequences in tooling — see the GitHub Security tab problem further down.

It is also not exhaustive. A scanner only tests what it reaches: routes behind a login, code paths a crawler never triggers, and flows that depend on business state stay untested unless something else covers them. Fluid Attacks’ documentation states that neither its scanner nor its pentesters support DOM APIs.

Why DAST matters

Runtime vulnerabilities are the ones a correct codebase can still carry. Deployment configuration, TLS settings, response headers, session handling and third-party integrations have no single line of source to inspect, and they only exist once the application is up.

Configuration fails where no source code lives

Two of Fluid Attacks’ own detection rules make the point. DAST-SCNRY reports a vulnerability “when the Content-Security-Policy header is completely missing from the HTTP response.” DAST-TNTER reads the CSP value and flags a missing script-src directive or unsafe configurations such as unsafe-inline or unsafe-eval. Both are classified as WEB DAST, with HTTP as target and HEADERS as technology.

Neither finding exists in the repository. They exist in the response.

Volume is not risk

Automated tools find most of the issues and a smaller share of the danger. In the State of Attacks 2026, Fluid Attacks’ automated tools identified 88.4% of all reported vulnerabilities but 55.8% of total risk exposure; manual testing accounted for the remaining 11.6% of vulnerabilities and 44.2% of exposure. For critical-severity vulnerabilities alone, 87% were identified through manual review. That is a division of labor, not a verdict on tooling: a scanner clears the field fast, and it is not what finds the issue that ends up on the incident report.

Applications are the target teams fix fastest

The same report counts 140,551 vulnerabilities in the “Application” target, carrying 3,903,585.51 CVSSF of risk exposure, and applications were the fastest target to remediate: at least 40% of their vulnerabilities were fixed within 30 days. Findings that arrive with a reproducible request tend to move, because a developer can replay them.

DAST at Fluid Attacks

Fluid Attacks runs Dynamic Application Security Testing as an automated scanner and Pentesting as a Service (PTaaS) as manual testing. The documentation states the split in two sentences: “DAST is an automated scanner. In PTaaS, pentesters test applications manually.” What follows is which surfaces each one reaches, how the scan is configured, and the two places where it stops.

What the scanner reaches, and where the pentester takes over

Attack surfaces covered by Fluid Attacks DAST and PTaaS

The scanner reaches seven surfaces; eight more need a pentester.

Attack surface

Automated DAST

PTaaS

DNS records

Yes

Yes

Headers

Yes

Yes

HTML content

Yes

Yes

SSL connections, encryption suites, protocols, X509 certificates

Yes

Yes

Unauthenticated HTTP endpoints

Yes

Yes

Webhooks

Yes

Yes

WebSockets

Yes

Yes

Authenticated HTTP endpoints

No

Yes

REST, SOAP, GraphQL and gRPC APIs

No

Yes

OpenAPI, Swagger and Postman specifications

No

Yes

DOM APIs

No

No

Source: Fluid Attacks’ supported attack surfaces. PTaaS is available exclusively in the Advanced plan.

The honest answer to “can DAST test APIs?” depends on which tool you mean. The web scanner ships as the fluidattacks/dast Docker container and covers the seven surfaces above. A separate tool, dast_api, ships as a distributed binary and “scans API endpoints declared in a Postman collection.” Fluid Attacks’ roadmap lists automated API DAST as in progress, so until it ships, API specifications are covered by pentesters.

What it looks like in practice

Element

How it works

Web scanner

fluidattacks/dast Docker container

API security testing

dast_api, a distributed binary that refuses to scan without a credential

Scope

dast.urls — “Each entry must have a url field. It can also have an optional environment_id field”

Connectivity check

dast.strict_connectivity, default false. When true, “the scanner exits with a non-zero status code if it cannot connect to a configured endpoint”

CI/CD pipeline

DAST GitHub Action, or the CLI in any Docker-capable runner

Output

SARIF or CSV, read as a workflow artifact

Build control

CI Gate, which breaks the build when findings violate your policy

Verification

Reattack, which re-tests a vulnerability after a fix attempt

Turn strict_connectivity on. Without it, a scan that never reached the target returns clean, which reads exactly like a scan that found nothing.

Why DAST findings do not fit the GitHub Security tab

Fluid Attacks’ documentation is blunt about it: “You cannot upload DAST findings to the GitHub Security tab. GitHub’s code scanning API requires a specific file and line number for each vulnerability. Web application vulnerabilities detected at runtime do not have this information.” Set output.format: SARIF or CSV and read the scanner’s output as a workflow artifact instead — worth knowing before you design a dashboard around a tab that will stay empty.

Authentication is where the scanner hands off

Top application weakness by mean risk exposure in 2026 report

Authentication absence or evasion leads, at 855.99 mean risk exposure.

Measure

Automated tools

Manual testing

Share of reported vulnerabilities

88.4%

11.6%

Share of total risk exposure

55.8%

44.2%

Share of critical-severity vulnerabilities identified

—

87%

Risk exposure from critical vulnerabilities detected

—

90%

Source: Fluid Attacks’ State of Attacks 2026, covering January 1 to December 31, 2025.

Put the two facts side by side. Automated DAST tests unauthenticated HTTP endpoints. The leading weakness in the Application target is authentication mechanism absence or evasion. The scanner confirms that a protected resource is in fact protected from an anonymous caller; it does not log in as two different users to check whether one can read the other’s records, which is what insecure object reference requires.

The benchmark puts a number on the same gap. Of the 1,201 vulnerabilities seeded in that test application, 73.2% required authentication to reach.

That is why PTaaS exists as a separate offering rather than a premium tier of the same scan.

Verifying the fix is part of the test

A report is not an outcome. Fluid Attacks’ documentation puts it directly: “As part of vulnerability management, examine the code again after each fix. A fix can add new vulnerabilities, or it can fail to fix the original vulnerability.” Reattack does that on request, and the surrounding practice moves the number more than the scan does: systems using the CI Gate reached a 72.2% remediation rate, against 58% for systems without it.

Go deeper: penetration testing as a service.

What a DAST scan can reach, and what it misses

A scan confirms a vulnerability in one of two ways. In-band testing reads the answer in the response; out-of-band testing waits for the server to call home.

[Image: db-dast-scnry-detection-strategy.png]

ALT (64): DAST detection rule for a missing Content Security Policy header

Caption: What the rule examines, and when it reports a finding.

What is tested

Example

What it confirms

HTTP headers

DAST-SCNRY

Content-Security-Policy is absent from the response

Script policy

DAST-TNTER

script-src missing, unsafe-inline or unsafe-eval

Authentication

Weakness 006

A protected resource answers an anonymous request

Authorization

Weakness 013

One user reads another user’s data via an identifier

Transport

SSL connections

Weak suites, outdated protocols, invalid certificates

In-band confirmation covers most of that list, and it breaks down on blind vulnerabilities: a server-side request forgery or a command injection that returns nothing to the caller leaves no evidence in the response body. Out-of-band DAST closes that gap with a payload that makes the target open a connection to a listener the tester controls. Fluid Attacks’ pentesters use Interactsh, “an open-source tool that finds vulnerabilities that cause out-of-band (OOB) interactions.”

In-band and out-of-band dynamic testing compared side by side

In-band confirms in the response; out-of-band confirms in a callback.Own work.

DAST for APIs

API security testing is where most introductory explanations get vague, so here is the precise version. A REST or GraphQL service exposes operations described in a specification, not links a crawler can follow, so DAST reaches them only if it is given that specification — a Postman collection, an OpenAPI document, or an introspection query.

The risks that matter there are authorization risks. The OWASP API Security Top 10 (2023) opens with API1:2023 Broken Object Level Authorization and also lists API2:2023 Broken Authentication and API5:2023 Broken Function Level Authorization. Testing for any of them means holding at least two authenticated identities and checking what each can reach — a configuration step, not a payload.

The 2018 Panera Bread leak shows what happens when nobody runs that test. As reported by Krebs on Security, customer records sat on the company’s website in plain text with no access control, and account IDs were sequential integers, so anyone could increment through the accounts. The flaw was reported on August 2, 2017 and was still live when the story published on April 2, 2018 — eight months. Krebs wrote that the incremental customer numbers suggested the count “may be higher than seven million,” and a later update put it at a figure that “appears to exceed 37 million.”

That case cuts both ways. An unauthenticated endpoint serving customer data is squarely inside what DAST reaches. The same flaw one login deeper would not have been.

How accurate is DAST?

DAST accuracy is four numbers, not an adjective: true positives, false positives, false negatives, and the share of the application actually covered. A vendor claim of “low false positives” says nothing about the third and fourth.

The reproducible way to compare tools is a fixed, deliberately vulnerable test suite. The OWASP Benchmark Project is “a Java and Python test suite designed to verify the speed and accuracy of vulnerability detection tools,” analyzable by static, dynamic and interactive tools alike, and every vulnerability it scores is exploitable. Running two scanners against the same suite and publishing both scorecards is evidence; a marketing page is not.

Synthetic suites have their own ceiling, and the gap shows in Fluid Attacks’ own numbers. In its benchmark of 36 third-party AppSec tools against an application carrying 1,201 vulnerabilities across 105 CWE categories and 461,500 CVSSF, a pentester found 1,076 of them — 89.6% recall at 100% precision. The best automated evaluator was Fluid Attacks’ own tool, covering SAST, SCA and DAST together, at 22.7% recall and 8.8% of total risk exposure. Among the other 34 tools, average recall was 1.7%.

A scanner that scores perfectly on a benchmark will still miss most of a real application, which has sessions, state and business rules that a test suite does not.

Automated DAST versus manual penetration testing

Automation covers breadth; people cover the cases that need a hypothesis. OWASP’s Developer Guide makes the same split on its DAST tools page: “The testing process can be carried out manually or be automated.” It adds that “usually business logic errors, race condition checks, and certain zero-day vulnerabilities can only be identified using manual assessments.”

The difference shows up at the payload level. A scanner sends a library of known patterns; a filter that rejects those patterns returns a clean result that is not a clean application. Jonathan Armas, Security analyst at Fluid Attacks, describes the failure mode in a walkthrough of bypassing SQL injection filters by hand:

“When the developers use blacklist filtering, as in the examples above, there is a risk of missing some parameter that can allow the injection.”

Finding that missed parameter means reading the filter’s behavior and building a payload for it. Three things a pentester configures that DAST does not infer: test users across every role, the business meaning of a transaction, and a chain of low-severity findings that together produce a high-severity one.

DAST, SAST and SCA are not alternatives

The three techniques look at different artifacts at different moments. Running one and calling the application tested is the most common mistake here.


DAST

SAST

SCA

What it examines

The running application

Source code, bytecode or binaries

Third-party components

Needs source code

No

Yes

Manifest and lockfiles

Needs a running app

Yes

No

No

Runs from

Test environment onward

First commit onward

First commit onward

Points to a line of code

No

Yes

In the dependency declaration

Finds configuration and deployment issues

Yes

Limited

No

Finds known CVEs in dependencies

Rarely

No

Yes

Static Application Security Testing (SAST) andSoftware Composition Analysis (SCA) run earlier and point at lines; DAST runs later and proves behavior. Secure Code Review and Reverse Engineering add the manual reading neither automated technique performs. Fluid Attacks offers SAST, SCA, SCR and RE as one all-in-one solution. Read more about how they combine, or launch the interactive demo.

How to run DAST in a CI/CD pipeline

Start by naming the targets. List every URL under dast.urls, give each one its environment_id when you track environments separately, and set strict_connectivity: true so an unreachable endpoint fails the job instead of returning an empty result.

Then decide what the pipeline does with findings. Point output.format at SARIF or CSV and publish the file as a workflow artifact, since the GitHub Security tab will not accept DAST results. Set a CI Gate policy that breaks the build on the severities you are not willing to deploy, and keep it narrow enough that developers do not learn to bypass it.

Finally, close the loop. Request a reattack after each fix and treat a vulnerability as resolved only when the re-test confirms it.

Run full scans against a staging environment that mirrors production, and keep production scans scoped and scheduled. An aggressive scan against a live transaction flow is a conversation to have with the application owner before the first run, not after.

Conclusions

DAST earns its place by testing the thing users actually reach: a running application, with its configuration, its headers, its transport and its exposed endpoints. The evidence it produces is reproducible, which is why these findings tend to get fixed quickly.

Its limits are just as concrete. DAST does not read code, does not reach what it cannot discover, and in Fluid Attacks’ implementation does not test authenticated endpoints or API specifications. The State of Attacks 2026 puts the consequence in numbers: automated tools found 88.4% of vulnerabilities and 55.8% of risk exposure, while 87% of critical-severity issues came from manual review. Treat the scan as the layer that clears the ground quickly, and plan for the testing that covers what it leaves standing.

Frequently asked questions

Does DAST need access to the source code?

No. DAST is a black-box technique that tests a running application from the outside. It sends requests to live endpoints and analyzes the responses, which makes it language-independent and usable against software you did not write. The trade-off: the finding describes insecure behavior without identifying the line of code responsible.

Can DAST test authenticated APIs?

It depends on the tool. Fluid Attacks’ automated scanner covers unauthenticated HTTP endpoints, DNS records, headers, HTML content, SSL connections, webhooks and WebSockets. Authenticated HTTP endpoints and REST, SOAP, GraphQL, gRPC, OpenAPI, Swagger and Postman specifications are covered by PTaaS, where pentesters authenticate to the application and follow API specifications. A separate binary, dast_api, scans endpoints declared in a Postman collection, and automated API DAST is listed as in progress on the roadmap. Neither technique supports DOM APIs.

How is DAST accuracy measured?

By four numbers: true positives, false positives, false negatives and coverage. A reproducible comparison needs a fixed test suite, such as the OWASP Benchmark Project, run against each tool with both scorecards published. Synthetic suites have a ceiling: in Fluid Attacks’ benchmark against a real application with 1,201 vulnerabilities, the best automated evaluator reached 22.7% recall and 8.8% of total risk exposure, while a pentester reached 89.6% recall at 100% precision. In that same application, 73.2% of the vulnerabilities required authentication to reach.

Is it safe to run DAST against a production application?

Usually, with scope control. Define the exact URLs to analyze, exclude destructive operations, and schedule intensive scans outside peak hours. Set strict_connectivity so an unreachable target fails loudly instead of returning an empty result. Full-intensity testing belongs in a staging environment that mirrors production.

Does Fluid Attacks’ DAST scanner replace a pentester?

No, and the company’s own data is the reason. In the State of Attacks 2026, 87% of critical-severity vulnerabilities were identified through manual review, and pentesters detected 90% of the risk exposure coming from critical issues. The scanner covers seven attack surfaces at speed; PTaaS adds authenticated endpoints and API specifications, where the most frequent application weakness of the year was concentrated.

Find what only shows up when the application is running

Our scanner tests your live endpoints on every pipeline run, and our pentesters take over at the authenticated surfaces and API specifications beyond it. Start free trial · Contact us

Get started with Fluid Attacks' application security solution right now

Reduce risk without slowing delivery

Fast and accurate results from one continuous security program powered by AI, scanners, and pentesters.

Reduce risk without slowing delivery

Fast and accurate results from one continuous security program powered by AI, scanners, and pentesters.

Reduce risk without slowing delivery

Fast and accurate results from one continuous security program powered by AI, scanners, and pentesters.