# Takumi AI Penetration Testing Now Supports Assumed-Breach Scenarios

Takumi AI Penetration Testing now supports assumed-breach testing.

Run a connection program on a machine that can reach your targets. Takumi treats that machine as compromised and, working through it, tests whether the objective you set can be achieved.

![Takumi AI Penetration Testing Now Supports Assumed-Breach Scenarios](/docs/_md-assets/36f5a28acf-eyecatch.png)

## Overview

Real attacks often begin with a single workstation compromised through a targeted phishing email or a malicious package, and spread inward from there. Until now, however, Takumi's penetration testing could only cover systems Takumi could reach directly. Testing anything else meant temporarily exposing it to the internet and restricting access with measures such as IP address allowlists.

With this release, you can treat a machine in your environment as compromised and run a penetration test through it. That machine can be a workstation on your office network or a virtual machine in a private cloud network, for example. Takumi autonomously explores the attack paths reachable through it and determines whether your objective can be achieved. Your targets never have to be exposed to the internet. And regardless of whether your perimeter holds, you can see how far the damage would spread once an attacker is inside.

Systems such as internal admin consoles and APIs reachable only from a private network can now be tested just like your internet-facing systems.

## How It Works

The connection program opens no listening port and connects to Takumi only through outbound HTTPS (port 443) from the starting machine, so no inbound port has to be opened on your side.

![How an assumed-breach test works](/docs/_md-assets/dc1c75833a-gateway-reach.svg)

Commands run on the starting machine go through the same safety checks as in any penetration test. Before running a command, Takumi checks it against the [scope](/docs/t/assessment/features/pentest#scope) and [destructive actions](/docs/t/assessment/features/pentest#destructive) policy you set, and stops it if it violates them.

## Getting Started

Assumed-breach testing is available from **Gateway settings** under **Assessments** > **Pentest** in the console. You create a penetration test almost the same way as before: in addition to the target and objective, you select the machine for Takumi to work through.

The starting machine can run on the following platforms.

- Linux (amd64, arm64)
- macOS (arm64)
- Windows (amd64, arm64)

On Windows and macOS, you can also enable desktop control, which lets Takumi read the screen and operate applications with the mouse and keyboard in addition to running commands.

For how to install the connection program and the detailed steps up to starting a penetration test, see the user guide.

▼ User Guide: [Assumed-Breach Tests](/docs/t/assessment/features/pentest/gateway/)