Penetration test
A penetration test points the Cerberon Agent at one or more hosts. Cerberon first runs a port and service scan, then the agent works from that output: it fingerprints each exposed service, chooses the tools that fit, tests it, and reports only what it can support with evidence. It needs an AI provider. See AI providers and privacy.
Start one from New Scan > Penetration Test.
How a test runs
Section titled “How a test runs”| Phase | What happens |
|---|---|
| Discovery | A port and service scan maps what each host exposes. Its output is saved with the scan and handed to the agent. |
| Service testing | The agent tests each service with the tools in the scanner image: port discovery, HTTP probing and fingerprinting, vulnerability template scans for known CVEs and misconfigurations, TLS configuration checks, web server checks and content discovery. |
| Proof | In a penetration test the agent may use exploitation tools, such as injection and credential testing tools, to prove a finding. In an assessment they are not run. |
| Reporting | Each finding carries the tool output behind it, a CVSS 3.1 vector where one can be scored, and a reproducible curl command where the finding came from an HTTP request a tool sent. |
A planned tool that did not run is listed in the report’s limitations with the reason, never treated as a clean result.
Test or assess
Section titled “Test or assess”| Option | Effect |
|---|---|
| Vulnerability Assessment Only on | Verifies issues without exploiting them. No denial of service, brute force, credential stuffing or data exfiltration. |
| Vulnerability Assessment Only off | A penetration test. Exploitation is permitted to prove a finding, and the report names the scan as a penetration test. |
Use Instructions for the tester to give the agent context, such as hosts to avoid or areas to focus on.
Targets
Section titled “Targets”- Up to 256 hosts per submission, one per line or comma-separated. Ranges are accepted.
- An optional Port on the Scan Details step.
- Loopback, link-local and cloud metadata addresses are refused.
For web applications and APIs with a structured method and per-check coverage, use Web and API testing instead.
Before you start
Section titled “Before you start”On the Review step you confirm that you are authorised to test every host in the submission. A penetration test reserves 3 GB of memory and has a default time limit of 2 hours, adjustable up to 48.