Honeypot Analysis Treated as Code with AI Assistance and Human Oversight
A novel approach treats honeypot analysis reports like code, using Git, pull requests, and AI drafting with mandatory human review to ensure accuracy and security.

Analyzing honeypot activity often results in a disorganized collection of terminal outputs and text files, lacking clear version control or accountability. To address this, security researcher Kevin Truong developed a system where honeypot reports are managed as code within a public Git repository, with all changes submitted via pull requests. This method ensures reproducibility, provides a clear audit trail, and enhances public safety by preventing the accidental leakage of sensitive information.
The workflow integrates an AI coding agent, Claude Code, which drafts the initial reports. However, to maintain human oversight and accountability, all AI-generated changes must be submitted as pull requests. These requests require a human approval before they can be merged into the main repository. This deliberate step acts as a critical gate, allowing for spot-checks of the content and ensuring that the AI's output is accurate and secure. The system also incorporates a public safety checklist and a pre-commit secret scanner to proactively identify and mitigate risks.
Key to this approach is the concept of reproducibility. By storing not only the findings but also the exact queries used to generate them, anyone can verify the data. This is crucial because the integrity of an indicator list or timeline depends entirely on the underlying query. The system also hashes evidence before it leaves the sensor, replacing sensitive information like IP addresses with placeholders, ensuring that even if the repository is public, critical infrastructure details remain protected.
The setup includes a separate GitHub account for the AI, which can propose changes but cannot approve or merge them. GitHub's ruleset enforces pull requests with at least one approval and blocks force-pushes, preventing the AI from unilaterally altering the codebase. Furthermore, a public-safety checklist is integrated into every pull request template, prompting the author to confirm that sensitive data has been scrubbed, placeholders are used correctly, and attachments have been manually reviewed.
This methodology offers significant benefits for various stakeholders. Students and new analysts gain a structured way to link claims to evidence and track the evolution of their analysis. Honeypot operators can publish findings with reduced risk of exposing their sensor's identity, thanks to the pre-merge scrubbing and checklist. Teams utilizing AI assistants can adopt a pattern that ensures AI-generated code is reviewed and controlled, clearly delineating where AI work ends and human intervention begins.
While version control is common for detection rules and infrastructure code, applying it to honeypot analysis transforms the process. It yields reproducible evidence, a reviewable history, and a secure method for leveraging AI drafters. The system's transparency, inherent in its Git-based structure, also highlights areas for improvement, with simple fixes often available within repository settings.
The system's effectiveness is demonstrated by the analysis of three specific attacks: an SSH brute force targeting MySQL credentials, a bot scanning for Telegram Desktop sessions and SMS modems, and a month-long campaign exploiting PHP with a potential PHP2Shell (CVE-2026-109618) component. Each of these investigations, detailed in separate folders within the repository, benefited from the rigorous code-like management and review process.
In conclusion, treating honeypot analysis as code, augmented by AI and governed by human review, provides a robust framework for generating trustworthy, reproducible, and secure security intelligence. This approach sets a precedent for how AI can be safely integrated into sensitive security workflows.