Skip to main content
Open a repository’s Runtime validation tab to test findings from a completed full scan against a running application. The tab is marked Early access. Use the evidence to decide which findings need remediation and to test the behavior again after you deploy a fix.

What It Does

Dynamic Testing exercises a configured running target with finding context, application profile inputs, encrypted runtime credentials, and, when needed, a compiled local application artifact. The goal is to produce evidence from the live application rather than relying only on static analysis. Runtime validation results use these verdicts: Disconfirmed findings can be moved into the false-positive workflow with runtime evidence explaining why the issue did not reproduce.

Vulnerability Classes

Dynamic tests can target runtime behaviors that are difficult to prove from source alone:
  • Prompt injection - tests whether AI-powered features execute injected instructions by manipulating inputs across user-facing and backend prompts. Findings show the injection path and model response.
  • IDOR - exercises object identifiers across authenticated roles to prove when users can access resources they do not own.
  • Vulnerability chaining - combines multiple runtime behaviors to show how smaller weaknesses become a practical attack path.
  • Out-of-band (OOB) vulnerabilities - detects blind SSRF, blind XXE, blind SQL injection, DNS exfiltration, and other vulnerabilities where the exploited behavior is not visible in the HTTP response. The validation agent generates unique callback URLs on oob.0path.ai and monitors for outbound requests from the target application, proving that the injected payload triggered a real server-side request.

Out-of-Band Testing and Network Requirements

OOB testing uses oob.0path.ai as the default callback server on ZeroPath Cloud. The validation agent injects unique URLs under this domain into the target application and monitors for DNS lookups, HTTP requests, and other protocol interactions that prove the vulnerability is exploitable. If your target application runs behind a firewall or egress filter, you may need to whitelist *.oob.0path.ai (DNS and HTTPS) so the target can reach the callback server. Without this, OOB tests will return no interactions even if the vulnerability is real. Self-hosted deployments do not use oob.0path.ai. They run a private callback collector inside the cluster on a dedicated DNS zone chosen by your platform team (for example oob.security.internal), and the applications under test must be able to resolve and reach that zone. See Self-hosted deployments below. A callback channel failure (DNS, TLS, or polling error) is always reported as an error, never as “no interactions”. The agent runs a positive control against the collector before and after each test, so a Disconfirmed verdict for an OOB finding means the callback channel was healthy and the payload still produced no interaction.

Prerequisites

Dynamic Testing requires:
  • A profile for the detected application. Choose a reachable deployed target, upload a compiled local artifact, or build the application from source.
  • The required inputs saved for that mode, such as the target URL and authentication fields for a remote target.
  • A completed, successful full scan. A selected-issue run also needs an eligible finding from that scan.
  • Start Scans on the repository to launch a run. Editing profile inputs also requires Edit SAST Configuration on the organization.
Generated input fields are represented as structured dashboard fields rather than a single JSON blob. Values are encrypted before they are stored, and readiness checks prevent validation from starting until required inputs are present. Local artifact validation currently supports Linux x86_64/amd64 compiled artifacts packaged as .zip, .tar, .tar.gz, or .tgz. The artifact runs inside ZeroPath’s Debian-based Runtime Validation gVisor container; other operating systems and CPU architectures are not supported.

Running Runtime Validation

Use this option to validate every detected application in a repository at once, without configuring individual application profiles first. ZeroPath builds and runs each application directly from source in an isolated container.
1

Open the repository

Go to the repository in ZeroPath and select the Runtime validation tab.
2

Auto-provision & Validate all applications

Click the Auto-provision & Validate all applications button in the toolbar at the top of the tab. This button appears when ZeroPath has detected applications in the repository and a completed full scan is available.
3

Monitor results

ZeroPath provisions a source-build profile for each detected application and queues a full validation run. Applications that are already running are skipped automatically. A confirmation message shows how many applications were provisioned and how many were skipped.
To use source builds automatically for eligible scans, enable Auto-provision Runtime Validation (build from source) under Full scans in scanner settings.

Reviewing Results

The repository Runtime validation tab shows recent validation runs. Selecting a run opens a dedicated run page with two top-level sections, selectable from the tab bar:

Issues

The Issues section shows validation results grouped by verdict:
  • Queued - issues still waiting for runtime validation.
  • Confirmed - issues proven exploitable at runtime.
  • Disconfirmed - issues that did not reproduce in the configured runtime environment.
  • Unable - issues where validation could not produce a reliable verdict.
Clicking an issue from any result tab opens the standard ZeroPath issue view, so the static finding, exploit walkthrough, remediation guidance, and runtime evidence can be reviewed together.

Exploit Chains

The Exploit Chains section shows multi-step attack paths discovered during the validation run — combinations of individual findings that, when chained together, form a practical exploit sequence. Use this view to understand how separately confirmed vulnerabilities can be combined by an attacker and to prioritize remediation of findings that contribute to high-impact chains. The table lists each chain with the following columns: Clicking a row expands it to show the full chain detail:
  • Impact — a narrative description of the real-world consequence if the chain is exploited.
  • Reasoning — the analysis behind why these vulnerabilities form a viable chain.
  • Attack steps — an ordered walkthrough of how an attacker would execute the full chain end to end.
  • Attack path — each individual step in the chain, linked to its originating issue, with its vulnerability class, severity, setup requirement, the role it plays in enabling the next step, and any configuration requirements.
  • Proof of concept — when available, a generated Python script that reproduces the exploit chain. You can copy the PoC to your clipboard using the copy button.
The active section is reflected in the page URL (?section=exploit-chains), so you can bookmark or share a direct link to either the Issues or Exploit Chains view for a specific run.

Issue-Level Evidence

When a finding has Runtime Validation output, the issue modal and full issue view include a Runtime validation section on the Evidence tab. It summarizes:
  • The latest verdict and run status.
  • What ZeroPath attempted during runtime testing.
  • Evidence observed during the attempt.
  • Why the finding was confirmed, disconfirmed, or marked unable.
  • A link to the dedicated run page for broader run context.
The section lives on the Evidence tab, so the Overview tab stays focused on the original finding details.

Structured Validation Data in the API

The issue detail and runtime validation API responses include two structured fields that give you machine-readable insight into what the validation agent did:
  • validationSteps — an ordered list of objects, one per action the agent performed. Each object contains:
    • target — the endpoint, workflow, object, or code path exercised (nullable).
    • action — the concrete action taken.
    • expectedResult — the expected safe behavior or exploit signal for this step (nullable).
    • observedResult — what the agent actually observed at runtime.
    • conclusion — why this observation matters for the verdict.
  • attackReproductionSteps — for confirmed findings only, an ordered list of natural-language steps that reproduce the confirmed attack. Suitable for inclusion in a bug report or remediation ticket.
Both fields are returned as arrays and are always present (empty when not applicable). The previously available agentGroupKey field has been removed from the API response.

Self-hosted deployments

Your platform team can enable execution on a supported self-hosted deployment with the Helm value runtimeValidation.enabled: true. Execution defaults to off. When it is off, you can still review and configure profiles. The repository page shows Runtime validation execution is disabled on this deployment. When enabled, everything runs inside your cluster:
  • Sandboxes run as short-lived pods under a sandbox RuntimeClass (gVisor by default) on nodes your team designates. Each run gets its own pod, and SOURCE_BUILD runs build and start the application inside that pod.
  • Callbacks go to a private collector reachable at a DNS zone your team delegates (for example oob.security.internal) instead of oob.0path.ai. Applications under test must use resolvers that forward that zone to the collector and must be able to reach it on ports 53, 80, and 443.
  • Network access from the sandbox is limited by NetworkPolicy to the private networks your team lists: the applications under test, the model proxy, internal package mirrors, and the collector. Public internet egress is not available from a sandbox, so REMOTE targets must be reachable on a private address and SOURCE_BUILD dependencies must come from internal mirrors configured by your team.
  • Local artifacts are staged on the shared scan volume and stored in your S3-compatible object store; they expire after 30 days.
Operational requirements, air-gapped package mirrors, and the acceptance checklist are documented in the self-hosted deployment guide (section DAST / Runtime Validation in private and air-gapped networks) delivered with each release.

Operational Notes

  • Full-application runs are blocked while another active run exists for the same scan and application profile.
  • Selected-issue runs are blocked only when a full-application run is active or when another active selected-issue run already targets the same finding.
  • Dynamic Testing depends on the configured deployed target, so a failure to reach the application should be treated as an environment or profile problem before treating the finding as disconfirmed.
  • For repositories with many scans, use the run list and result tabs instead of scanning issue-by-issue from the main issues table.