The network, as the attacker sees it.

AttackBase runs your penetration test with AI agents inside hard guardrails, replays every finding before it counts, and writes the report from the evidence.

Live view of a sample engagement
How it works

Plan, run and prove the test in one place.

AttackBase covers the whole engagement, from the rules of engagement to the signed-off report.

01 · PLAN

Scope in the dashboard

Targets, exclusions, testing window, detection posture and rules of engagement live in one place, and the platform enforces them.

  • In-scope and excluded ranges
  • Testing windows and blackouts
  • Quiet, normal or loud posture
02 · RUN

Agents do the legwork

Recon, web, Active Directory, cloud and exploitation agents work the scope in parallel. You watch it happen on a live map.

  • Every action logged with its reason
  • MITRE ATT&CK mapped from a reviewed table
  • Isolated runners for external scans
03 · PROVE

Only what reproduces

A finding counts only after an independent replay reproduces it. Then it goes into the report with its evidence attached.

  • One-click re-verify per finding
  • Screenshots over console dumps
  • Retest the whole report later
Replay

Every finding replays the way it was found.

Each finding carries the exact recipe that found it. AttackBase replays it on its own before the finding can count, and you can replay it again any time, during the test or months later for the retest.

  • Independent replayA separate verifier, not the agent that found it, reproduces the finding.
  • Traffic-backedRequests and responses are recorded, so the proof is the real exchange.
  • Export and rerunExport an engagement with its logs and evidence, import it later and rerun every validation.
F-07 · replay · step 3 of 5
$ verify F-07
recipe sha256 e07f22…1c4d · tools pinned
[*] relaying coerced login to /certsrv
[*] authenticating as CORP/DC02$
[+] certificate issued, ID 118
originalsha256 e07f22…1c4d
replaysha256 e07f22…1c4d
● Reproduced exactly. Finding counts.
Guardrails

Rules of engagement, enforced in code.

Agents move fast, so the limits you agree with the client are checked in code before every action instead of sitting in a document.

SCOPE

Scope gate

Every target is checked against the scope and exclusions before a command runs. Out of scope is blocked, not warned.

TIME

Testing window

Nothing touches a client system outside the agreed window or during a blackout. Retest windows are set separately.

ISOLATION

Locked-down runners

External scans run in short-lived containers that can reach only the internet, never anything internal, and are torn down after each scan.

Reports

The report writes itself from the evidence.

Findings arrive with screenshots, the commands that proved them, CVSS, CWE and ATT&CK already attached. Rules catch the corrections you'd otherwise make by hand, every time.

EVIDENCE

Evidence packs

Each finding gets a screenshot-ready evidence page, redacted the way your report rules say.

MAPPING

Standards built in

CVE, CWE and MITRE ATT&CK on every finding, with an appendix of techniques used.

RETEST

Retest in a click

When the client says it's fixed, replay every finding and hand back what changed.

Get started

See AttackBase on a real engagement.

We'll walk you through a live test from scope to report and show you a finding replayed on demand.