PentestGPT

Visit Website

Introduction

PentestGPT started as a research project, published at USENIX Security 2024, that used LLMs to guide human penetration testers. The current version is an autonomous agent that drives Claude Code or Codex to run tools and reason without human intervention.

It works through a multi-stage pipeline in which each stage feeds the next. In CTF mode the stages are recon, exploit and walkthrough; in pentest mode they are asset discovery, vulnerability identification and report. It supports web, crypto, reversing, forensics, PWN and privilege-escalation challenges, shows live activity as it works, and can save and resume sessions.

You run it against a target with optional instructions, for example a WordPress site with a focus on plugin vulnerabilities. A Docker image bundles the tool with both Claude Code and Codex CLIs; a one-time login stores sessions in named volumes, and make docker-run selects the target, backend, model and mode.

The original human-in-the-loop design is preserved as pentestgpt-legacy, which supports a wider set of providers: OpenAI, Anthropic, Google Gemini, DeepSeek, xAI, Qwen, Moonshot and local Ollama models.

As with any offensive tool, use it only on systems and labs you are authorised to test; CTF platforms and your own lab machines are the natural starting point.

The source code is at GreyDGL/PentestGPT on GitHub under the MIT licence, with about 16,400 stars; the most recent commit was on 14 July 2026.

PentestGPT is best understood as a research-grade agent rather than a turnkey enterprise product. Its staged pipeline makes the agent’s reasoning visible: you can see what it discovered during reconnaissance, which vulnerability it chose to pursue and why, and how it moved from one step to the next. That transparency is valuable for learning and for judging whether an AI approach is sound.

A good way to evaluate it is on retired capture-the-flag machines or a deliberately vulnerable lab application such as an intentionally insecure web app. Compare its progress with what an experienced tester would do in the same time, note where it gets stuck, and decide which phases, such as reconnaissance or report drafting, it can usefully speed up in your own work.

Model choice matters. Because PentestGPT drives Claude Code or Codex, its capability and cost track those backends, and the legacy mode lets you try other providers. Keep logs of each session so results can be reviewed and reproduced, and never point it at systems outside an agreed scope.

Finally, treat its reports as drafts. An autonomous agent can summarise what it did and what it found, but a human should confirm severity, check for false positives and write the remediation advice that a development team will actually act on. Used that way, PentestGPT saves time on the repetitive parts of an engagement without lowering the quality of the final report.

For teams teaching security, PentestGPT is also a useful demonstration tool. Watching an agent work through a lab machine, and discussing where its reasoning was right or wrong, is a practical way to help junior testers understand methodology, while reinforcing that authorisation and scope come first in every engagement.

Used carefully, it is a strong teaching and research companion.

Features

1. Autonomous pipeline: Staged phases feeding each other.
2. CTF and pentest modes: Different pipelines for each.
3. Claude Code or Codex backends: Pluggable agent backends.
4. Challenge categories: Web, crypto, reversing, forensics, PWN and privesc.
5. Live walkthrough: Real-time progress tracking.
6. Session persistence: Save and resume work.
7. Docker bundle: Both CLIs packaged with persistent logins.
8. Legacy mode: Human-in-the-loop with many model providers.

Pros

1. Research pedigree: Grounded in a peer-reviewed paper.
2. Great for learning: Shows its reasoning step by step.
3. Free to use: Open source under the MIT licence, so the core tool costs nothing to use.
4. Actively maintained: The repository had commits as recent as 14 July 2026.
5. Community: About 16,400 GitHub stars.

Cons

1. CLI subscriptions needed: Claude Code or Codex must be installed and authenticated.
2. CTF-leaning: Strongest on lab-style targets.
3. Ethics and authorisation: Offensive use requires permission.

Usage and Links (Example)

pentestgpt --target 10.10.11.234
pentestgpt --target 10.10.11.50 --instruction "WordPress site, focus on plugin vulnerabilities"

# Docker flow
make docker-login
make docker-run TARGET=10.10.11.234 BACKEND=claude MODEL=opus MODE=pentest

Resources:

• PentestGPT website

Pricing

PentestGPT is free and open source under the MIT licence.

Recommended for?

1. Security learners and CTF players.
2. Red teams prototyping AI-assisted workflows in labs.

Conclusion

PentestGPT is a well-known open-source agent for autonomous pentesting and CTFs.

Treat it as a lab and research tool before using it on real engagements.

On CheckOps, PentestGPT is filed under the Security testing stage. Related tools listed here include Metasploit, Kali Linux, Burp Suite, Nmap, SQLMap and Nuclei. Which one fits best depends mostly on your stack, your team’s size and skills, and whether you prefer to run tools yourself or pay for a managed service. This page reflects the official website, documentation and repository as they stood on 11 November 2026.