Authorized Testing
Effective date: 2 October 2026
Some Enji Guard features — runtime checks and Auto-pentest among them — send active requests to the websites, APIs and web resources you link. Because those requests reach a live target, they run only with your explicit consent. This page sets out the conditions: who can authorise a test, what is in scope, what may happen, and what is prohibited.
Authorisation requirement
You may run active checks only against systems that you own or have written permission to test. By enabling active checks, you confirm that:
- You own the target or have written authorisation to test it.
- Your organisation permits the test.
- Your hosting or infrastructure provider permits the test.
- The target is in scope.
- You understand that active checks may affect the target.
Consent validity
Consent applies to a specific project, website or domain, and selected scope. A renewed confirmation is required if:
- The website, domain, or target scope changes.
- The project is transferred to another owner or organisation.
- Ownership or authorisation may have changed.
- Materially more aggressive active-test types are added to the scope.
- A repository, website, or project is disconnected and reconnected.
- Twelve months have passed since the previous confirmation.
What Enji Guard may do during active testing
Within the approved scope, Enji Guard may discover endpoints, forms, parameters, and headers; send common vulnerability payloads; test for issues such as SQL injection, cross-site scripting, SSRF, IDOR/BOLA, broken access control, authentication bypass, and sensitive data exposure; generate the normal request volume needed to exercise endpoints; record reproduction steps; and generate a report.
Possible impact
Active checks can cause temporary errors or slowdown, higher request volume, security alerts and WAF or log noise, temporary account lockouts, and the exposure of bugs that need urgent remediation.
You accept these risks for the targets you authorise, and you are responsible for backups and for monitoring the target during testing. There is no way for you to stop a run once it has started — we can stop one if you ask, but by email rather than at the press of a button. See Stopping a test below.
Prohibited testing
- Denial-of-service or load testing.
- Credential stuffing, password spraying, or brute force.
- Malware, persistence, lateral movement, or destructive payloads.
- Data exfiltration.
- Privilege escalation outside the agreed scope.
- Scanning unrelated domains, IP ranges, tenants, customers, or shared infrastructure.
- Testing targets prohibited by a hosting, cloud, or customer contract.
What a test may reach, and why that is a data protection question
A penetration test works by getting at things it should not be able to get at. Checks for SQL injection, broken access control and IDOR succeed precisely when they reach data the tester was not meant to see. On a live target, that data is your users’ personal data — not yours, and not ours.
So before you authorise a test, three things are yours to settle.
- You need a lawful basis for us processing your users’ data in the course of the test, and it is not covered by the basis you have for running the service they use. Security testing is usually defensible on legitimate interests, but it is your assessment to make and record.
- Consider whether a DPIA is required. Testing that systematically probes access controls on a production system holding personal data is the kind of processing that often triggers Article 35. We are not telling you it does; we are telling you not to skip the question.
- Prefer a staging environment with synthetic data. Everything above becomes easier if the target is not carrying real people’s records. Where a production test is unavoidable, scope it as narrowly as the finding you are hunting.
What we do with what a test reaches.
- We record only what is needed to prove and reproduce the finding. A report shows that a record could be retrieved, not the contents of the records.
- We do not harvest. Nothing extracts data beyond the sample that evidences the weakness.
- If we hit something we clearly should not have — special category data, credentials belonging to a third party, another tenant’s records — we stop that check, tell you, and delete what we captured unless you need it retained to reproduce the issue.
- Special category data is never sought. It can still be reached by accident, which is what the previous point is for.
- Access to these reports is restricted to one named person on our side, not the team. See DPA Annex 2, Who can see what, which sets out human access in full.
Handling reports
Auto-pentest reports may contain sensitive vulnerability details. Treat them as confidential, and do not publish or share a report unless you have authority to disclose the target, findings, reproduction steps, and remediation information.
The Service’s publication controls reach these reports too. They default to private and we never use them on your behalf — Terms of Service section 13 sets out how they work and what revoking a link does and does not undo. A report describing an unfixed weakness in a live target is the last thing that should be published by accident.
Stopping a test
Disabling the job stops anything further from starting. It does not touch a run already under way.
A run under way can be stopped, but only by us and only by hand. Write to contact@enji.ai and we will stop it. That route is a mailbox, not a control in the product, so how fast it happens depends on someone reading the message. Do not plan on it as an emergency brake.
Scope a test on the assumption that it will run to the end. If a run against a particular target could do damage you would need halted within minutes, that target is one to test in staging with synthetic data instead — see What a test may reach above.
If a run fails, or we stop one at your request, the credits reserved for it are returned to your balance. See Terms of Service for how credits are reserved and spent.
Consent records
Enji Guard may store consent records showing that a user confirmed the active-testing conditions for a target. Consent records may survive disabling and re-enabling a job, and are kept even after a consent is no longer valid for future checks — for 3 years after it stops being valid, because the record is the evidence that someone authorised probing a live system. See Data Retention & Deletion.
Enji Guard