# Incursion: the AI agent behind our pentests

Source: https://aipentestservices.com/incursion/
Publisher: Excession Labs (aipentestservices.com)
Last updated: 2026-11-09

Incursion is the terminal AI agent behind our pentests. See how its coordinator, workers, tool rules and evidence standard work in authorized testing.

> **Short answer:** Incursion is a terminal AI agent for authorized security testing, built by Excession Labs. A coordinator splits an engagement into tasks and several workers run them in parallel under allow, ask and deny tool rules. A task is done only when each finding links to its evidence. This service uses the same agent, operated by Excession Labs.

Incursion is a terminal AI agent for authorized security testing. Excession Labs builds it, and our pentest service uses the same agent, operated by Excession Labs for you. The product page at [excessionlabs.ai/incursion](https://excessionlabs.ai/incursion/) has the install and the full description.

## What is Incursion?

Incursion is an agent that works in a terminal. You name a system you are permitted to test, and it divides the job into tasks. Several workers then run those tasks at the same time.

Operators use it for:

- reconnaissance
- web and API testing
- code audit
- exploit development
- reverse engineering
- malware analysis

This service uses only the web and API testing.

## How does it split the work?

A planner, called the coordinator, turns the engagement into objectives and then into tasks. Workers take tasks and run them in parallel, so one agent can cover many endpoints and roles in one test.

Each worker keeps its own notes and findings. They stay after a run, so they can be reviewed and retested. In a service engagement, Excession Labs is the operator.

## How do the tool rules work?

Each tool has one of three rules, and the operator sets them:

| Rule | Behavior |
|---|---|
| Allow | The tool runs on its own |
| Ask | The tool waits for approval |
| Deny | The tool is blocked |

The rules keep the agent inside the written scope. The [safety page](/safety/) describes how we set them in an engagement.

## What counts as a finished task?

A task is done only when each finding links to its evidence. Each finding includes the evidence, the steps to reproduce it and a suggested fix. A claim without evidence is not a finding.

Incursion has its own harness: an agent loop, its own tools and session history. It supports retesting a finding, so a fix can be checked with the original steps.

## What does Excession Labs add in the service?

The agent does the testing work. Excession Labs:

1. Agrees the scope with you in writing.
2. Operates the agent and sets the tool rules from the scope.
3. Checks the evidence for each finding.
4. Writes the report.

You receive the report in HTML, PDF and Markdown, with proof-of-concept scripts. [How it works](/how-it-works/) covers each step, and [AI vs manual pentesting](/ai-vs-manual-pentesting/) covers the split between agent and people.

## What does Incursion not do?

Incursion does not know your business rules unless you explain them. It does not find all vulnerabilities, and we do not call a system secure after a test. The agent tests only what the written scope allows. The report lists limitations and open questions, so you can see what the test did not cover.

## Where can I read more about agentic testing?

[How agentic pentesting works](/guides/how-agentic-pentesting-works/) covers the architecture in general terms: planner, workers, tool use, memory and evidence. To try the service on a small scope, apply for the [free pentest pilot](/pilot/).

## Questions

### Is the service a different product from Incursion?

No. The service uses the same agent that Excession Labs publishes. The difference is who operates it: for the service, Excession Labs agrees the scope, runs the agent, checks the evidence and writes the report.

### Can I run Incursion myself?

Incursion has a public install, described on the Excession Labs site. Pentesters, red teams and bug bounty hunters use it for authorized work. This service is for teams that want Excession Labs to run the test and deliver a report.

### Does Incursion work without human control?

No. The operator sets the tool rules and approves actions that wait for approval. We do not call it autonomous without human control.

### Does Incursion find every vulnerability?

No. We do not claim that. It reports findings that it can back with evidence, and the report lists limitations and open questions.

## Sources

- [Incursion, Excession Labs](https://excessionlabs.ai/incursion/)
- [About Excession Labs](https://excessionlabs.ai/about/)