An AI Full Pentest is a deep, on-demand penetration test of a web application, run end-to-end by Intruder's autonomous AI pentesting agents. It gives you the depth of a manual pentest, on demand: time to exploit has gone from months to hours, and annual or quarterly pentests can't keep up, so you can test with every release. It covers your web apps specifically — not your whole external network — in a fraction of the time and cost of a manual engagement.

It's code-assisted: the agents read your source to see how the app is built and map your APIs from the code (no schema to upload), then actively test the live app — using the same methods as experienced pentesters — to discover new vulnerabilities, not just validate known ones. Every finding is proven exploitable, with live evidence and the exact source file and line where it originates.

That reach is the point: it regularly surfaces critical issues that scanners and human testers both miss — authentication and authorization bypasses, cross-tenant access to another customer's data, and business-logic flaws with direct financial impact. Scanners flag what they can detect but can't tell you what's a real risk in your environment; the agents confirm each finding against the live app, so you get fewer rabbit holes and faster remediation. When it's done, you get a detailed PDF report suitable for compliance evidence under SOC 2, ISO 27001, and PCI DSS.

An AI Full Pentest is separate from AI Issue Triage & Investigation, the credit-based feature that validates issues you've already detected; a Full Pentest tests a whole application from scratch and is a one-off purchase. You set one up in a short guided flow — scope, source code, optional credentials and context, then pay — and your progress saves as you go. For how it differs from scanning, see AI Pentest Features in Intruder: Overview.

To start, go to Pentests, begin a new pentest, and choose AI pentest. The setup opens with a progress bar and a tracker down the side so you can see what's left. Everything you enter saves automatically — close the tab and come back, and your setup is still there.

Two parts are required (Scope and Source), and two are optional (Accounts and Context).

Scope tells the AI agent what to test.

Test accounts let the agent log in and test the authenticated parts of your app, which is usually where the interesting risk lives. If your app has any kind of login, adding credentials matters: without them, the agent can only test what's reachable when logged out, so most of the application goes untested. Only skip this step if the target is a fully public site with no authenticated area.

Add as many accounts as you can, covering every access level in your app — and, if it's multi-tenant, accounts belonging to different tenants or customers. This lets the agent test for privilege escalation and cross-tenant authorization failures, where a user reaches data or actions they shouldn't, which are among the most common critical issues it finds. As a rule of thumb, more is better — we test our own platform with roughly a dozen accounts covering every access level.

To add an account:

Source code is where you connect the repository that the agent reviews. This is a required part of an AI Full Pentest — the test is code-assisted, so the agent needs to read your code to pinpoint where issues originate.

Repositories are connected read-only, are not accessed by a person, and are deleted once the pentest completes.

Context is where you point the agent at what matters. Every field is optional, but the more you give, the more targeted the results.

Checkout shows a summary of your configuration to review, then handles payment.

To discard your progress and start fresh, open the ⋯ menu at the top of the setup and select Start again. This menu is available throughout setup. Starting again clears your saved draft completely — there's no undo.

After you submit, you're taken to the pentest's status page, where the AI agent's progress is shown in real time — moving from Pending to Running to Completed. You can leave and return; the pentest keeps running regardless.

Before testing gets fully underway, Intruder runs preflight checks that confirm it can access your repository and log in with any credentials you supplied. If a check fails, you'll see a clear error — "Source code check failed" or "Accounts check failed" — with a Contact us button, so a misconfigured repo or credential doesn't quietly waste a test.

You can cancel at any point up until a report has been generated — while the pentest is still running, a Cancel pentest button is available on the status page. Canceling stops further testing, but can't undo actions the agent has already taken. It does not refund the test, and it won't produce a report for that engagement. If you want to resume a canceled pentest, contact us — but we can't guarantee it can be continued. Once the report exists, the pentest is complete and can't be canceled, but you can still leave feedback or raise questions with support.

The deliverable is a downloadable PDF report, accessed through a secure, time-limited link, and built to be handed to a third party. It follows a consistent structure:

Findings come with built-in recommended remediation SLAs — Critical within 24 hours, High within 7 days, Medium within 30 days, Low by the next release — so you have a timeline you can commit to internally or show an auditor.

You can use the report internally and share it with your own customers, regulators, auditors, and professional advisors to demonstrate your security posture. You can't share it with other third parties — in particular, providers of competing security testing services — without Intruder's written consent.

Yes. An AI Full Pentest is code-assisted, so a connectable GitHub or GitLab repository is required — the agent reviews your source as part of the test. If you can't connect a repo, you can't run one yet.

Test accounts are optional in the wizard, but if any part of your app sits behind a login you should add them — without credentials the agent can't reach those areas, and most of the app goes untested. For a site with no authenticated area, you can skip this.

It's a web application penetration test, run both unauthenticated and (if you supply credentials) authenticated. It covers the OWASP Top 10 and beyond — authentication and session handling, authorization, business logic, input validation and injection (XSS, SQL injection, command injection, and similar), file uploads, information leakage, and server configuration. Denial-of-service, social engineering, phishing, and physical security testing are out of scope.

Each AI Full Pentest is a one-off, per-engagement payment, separate from your AI pentest credits. The exact price is shown at checkout before you pay; it varies by currency and by whether you're on a paid plan, and existing paying customers and early adopters currently get a discount.

No. AI pentest credits are the monthly allowance used for AI Issue Validation (shown in the portal under Triage & investigation). An AI Full Pentest is billed separately as a one-off payment per pentest.

Most run in minutes to hours, depending on the size and complexity of your app and how much context you've given the agents. You don't need to wait around — progress shows on the status page in real time, and you can leave and come back while it runs.

You can, up until a report has been generated, using the Cancel pentest button on the status page. However, canceling stops the test, but it isn't refunded, and won't produce a report for that engagement. Resuming a canceled pentest isn't always possible, so contact us if you need to. Once the report exists, the pentest is complete.

Credentials are encrypted at rest and never included in reports or logs. Source code and anything else you provide is processed only inside the engagement's isolated testing container, deleted after the engagement, and never accessed as a complete repository copy by Intruder staff. Your data is never used to train AI models.

For the exact retention periods, sub-processors, and data flows, see the AI Pentesting terms and our Intruder Trust Center - AI FAQs.

The methodology is distilled from Intruder's in-house security team, whose qualifications include CREST certifications and the Offensive Security Certified Professional (OSCP). Intruder is a CREST member company, is SOC 2 Type II compliant, holds Cyber Essentials, and was selected for the UK government's GCHQ Cyber Accelerator. The agent applies the same methods as a human tester — sending requests, analyzing responses, and validating findings against the live app. Full details are in the About Intruder appendix of every report.

No. Your data, including any source code, is never used to train or fine-tune any AI model, and each customer's data is kept isolated from every other customer's. The Intruder Trust Centre has the full breakdown of what's sent where.

An AI Full Pentest is a one-off engagement that produces a single report. You can't re-run it, extend it, or change the target after it launches without buying another. A purchased pentest must be used within 12 months.