Yeti-791/Awesome-Offensive-AI-Agentic-Landscape

This document curates open-source projects, academic papers, capability benchmarks, and commercial solutions (international & China) in AI penetration testing, LLM red teaming, autonomous offensive agents, and vulnerability discovery—aimed at helping researchers, security engineers, and enterprise decision-makers quickly form a holistic view.

What it solves

This repository is a comprehensive, curated index of the offensive AI security landscape. It helps researchers, security engineers, and enterprise decision-makers quickly find and understand the key open-source projects, models, skills, MCP servers, academic papers, benchmarks, and commercial solutions in AI-driven penetration testing and autonomous red-team agents. Instead of searching across many sources, you get an organized overview with counts, star ratings, and links.

How it works

The README is structured into clear sections. It lists 46 open-source agent projects (spanning pentesting, red-team, CTF, etc.) sorted by GitHub stars, 11 offensive/safety-specialized open-weight models (split into uncensored offensive models and security-aligned models), 3 pure Agent Skill resources, 7 security-tool MCP servers, 116 academic papers, 13 benchmarks, 9 awesome lists, and 32 commercial solutions (20 international and 12 from China). It also highlights notable competitions like DARPA AIxCC. The data is maintained with notes on updates, star conversions, and pruning decisions.

Who it's for

This is primarily for security researchers, penetration testers, red-team operators, and enterprise security decision-makers who want to explore or adopt AI-driven offensive security tools. It is also useful for academics tracking the latest research (papers and benchmarks) and for practitioners selecting models, MCP servers, or commercial products for their work.

Highlights

  • Top open-source agents include strix (60.1k stars), shannon (47.6k), pentagi (22.2k), and PentestGPT (15.2k).
  • Distinguishes truly uncensored offensive models (e.g., Qwythos, DeepHat/WhiteRabbitNeo, BaronLLM) from security-specialized but aligned models (e.g., VulnLLM-R, Meta-SecAlign).
  • Lists MCP servers that bridge AI agents to real tools like Burp, Metasploit, and Shodan.
  • Includes 116 curated papers (2023→2026) and 13 benchmarks for evaluating agent capabilities.
  • Documents DARPA AIxCC 2025 finalists, showcasing state-of-the-art cyber reasoning systems.

Related

  • Project
    Mrjie7205/serenity-bottleneck-hunter

    A disciplined investment research skill for AI agents that reverse-maps supply chains to identify overlooked upstream bottleneck stocks based on a specific trading methodology.

  • Project
    DeHor-Labs/mcp-fiscal-brasil

    A Python MCP server that gives AI assistants and other tools instant, offline‑first access to Brazil’s fiscal data (CNPJ, NF‑e, SPED, eSocial, Simples Nacional, 2026 tax reform). No API keys needed; optional premium source adds real‑time official data.

  • Project
    ohdearquant/lionagi

    A governed multi-agent orchestration framework for building typed, inspectable, and durable agent workflows via Python or a powerful CLI.

  • Project
    GizClaw/flowcraft

    A modular Go toolkit for building provider-agnostic AI applications with support for long-term memory, durable agent execution, and MCP-based plugins.