hzysvilla/agent4pentest_survey

Agent4Pentest_Survey is a curated reading list and companion repository for an academic survey paper about how AI language models are being used to automate penetration testing, which is the practice of testing computer systems for security weaknesses. The repository itself is not a testing tool. It is a reference collection of 81 research papers published between 2023 and mid 2026, organized so researchers can find related work easily and contribute new papers through pull requests. The survey organizes this research into a four phase timeline. It starts with early systems where the AI model only reasoned about what to do while a human carried out every action, moves through single AI agents that could directly run scanning and exploit tools themselves, then to systems where multiple specialized AI agents split up the work under a coordinator, and finally to newer systems that learn from verifiable outcomes like successfully capturing a flag in a competition or gaining system access. Papers are also grouped into six categories covering evaluation benchmarks and test environments, general purpose autonomous testing systems, tools built for narrow attack scenarios, systems focused on capture the flag style challenges, defense focused research, and other survey papers that summarize the field. Each entry lists the paper title, a link to the paper, and a link to its code repository when one exists, along with information about where it was published. This repository is meant for researchers, students, and security professionals who want an organized overview of academic progress in AI driven penetration testing, not a ready to use hacking tool. It is actively maintained and welcomes contributions of new papers, benchmarks, and datasets.