Skip to content

Write Pentest Reports with AI

Manual penetration testing produces the findings that scanners cannot. It also produces the writing nobody has time for. Faction's AI report writing assistant closes that gap without handing the report to a chatbot: the tester records the evidence, and prompts that your team wrote turn it into finished prose in your voice.

Evidence in, report text out

The workflow follows how testers actually work. While exploiting an issue you jot down the steps to reproduce, the request, the response, the screenshot. That is the input. From it, Faction generates:

  • A vulnerability description written from your details, not a generic paragraph about the vulnerability class, in terms the developer can act on, with references from OWASP, CWE and NIST found by web search where they exist.
  • A recommendation grounded in what was observed rather than a boilerplate fix.
  • A finding title suggested from the description and details.
  • An executive summary for the whole assessment: an overview, every finding summarised in severity order, and a prioritised remediation plan.

Each is one click on the editor toolbar. See AI Content Generation for the walkthrough.

Your voice, not the model's

Every prompt is written and owned by an administrator. Length, structure, tone, whether to use headings, whether to add references, what to leave out: all of it is in the prompt, so the output matches how your team writes and never has to be edited into shape. Change the prompt once and every report that follows changes with it. Faction ships with three prompts to start from, and you can add your own for any field on an assessment or a finding.

A free-form Ask AI box is there for the rest: "rewrite this in simpler language", "summarise the details field", "expand this with remediation steps".

The model sees the assessment, and only the assessment

Prompts do not run on a blank page. The model is given tools to read the assessment it is running in, its details and its vulnerabilities, so an executive summary covers every finding without anyone listing them. Those tools are scoped on the server to that one assessment. The model cannot name another assessment and cannot reach data outside the one you are in.

Bring your own provider, keep your data

Faction connects to OpenAI, Anthropic, Azure OpenAI, OpenRouter, or any OpenAI-compatible endpoint, including models you host yourself. There is no Faction-run AI service in the middle. Each prompt can pin its own provider and model, so a description that searches the web can use a larger model while an executive summary uses a faster, cheaper one.

Before any text is sent, Data Privacy masks API keys, tokens, private keys, credentials and email addresses with placeholders and restores the real values in the output. Point it at a self-hosted Presidio instance to mask names, phone numbers and other personal data as well. If the masking service is configured but unreachable, generation is blocked rather than sending unmasked text. See AI Configuration.

Watch it

The Faction 2 AI prompts video walks through a finding from steps to reproduce to a generated report in a few minutes.

Try it

AI report writing is part of the open source edition. Run Faction, add a provider API key under AI Configuration, and the AI buttons appear on every editor.