adversarial-ai 开源项目推荐（94 个）

qybaihe /bla

BLA（Blind LLM Agents）——对抗性防御盾，保护二进制文件、Electron 应用、JS 包、Python 和 WebAssembly 免受 REA 等 AI 逆向工程智能体的攻击。Poison -> Exhaust -> Hallucinate。

codexstar69 /bug-hunter

对抗性人工智能错误猎人，具有Claude Code、Cursor、Codex CLI、GitHub Copilot CLI、Kiro CLI、Opencode、Pi Coding Agent等的自动修复技能。多代理管道会发现安全漏洞、逻辑错误和运行时错误，然后在安全分支上自主修复它们。

intent-driven-dev /intent-driven-template

面向意图驱动开发的 OpenSpec 和 OpenCode 模板，包含规范、ADR、C4 图、Gherkin、TDD、多模型对抗式规范编写、领域术语表以及可复用工程工作流。

regaan /basilisk

Basilisk — Open-source AI red teaming framework with genetic prompt evolution. Automated LLM security testing for GPT-4, Claude, Grok, Gemini. OWASP LLM Top 10 coverage. 32 attack modules.

SnailSploit /AATMF-Adversarial-AI-Threat-Modeling-Framework

AATMF | An Open Source - Adversarial AI Threat Modeling Framework

fallen-angel-systems /fas-judgement-oss

Open-source prompt injection attack console. Test AI security by firing categorized attacks at any endpoint.

karloks2005 /JailbreakLab

Test and evaluate Large Language Models against prompt injections, jailbreaks, and adversarial attacks with a web-based interactive lab.

obscuralabs-AI /Symbolic-Prompt-PenTest

Semantic Stealth Attacks & Symbolic Prompt Red Teaming on GPT and other LLMs.

K4N3CO /JailBreak-Ai

JailBreak-Ai is an advanced AI red-teaming app for Android. It generates high-fidelity payload prompts to test AI safety filters, model robustness, and logical boundary security for Jailbreaking capabillities.

TheApexWu /crucible

1st Prize (General Judge), Self Improving Agents Hack at Datadog. Two identical LLM agents play Split or Steal; neither is told to betray the other, and deception develops over the game. Built with @evancorrea.

jasminemoreira /Versus

Adversarial verification layer for AI coding assistants. Based on IACDM — Interactive Adversarial Convergence Development Methodology. The AI proposes. Versus critiques.

Cogensec /agentegrity

The open standard for AI agent integrity. Evaluate, enforce, and prove that autonomous agents are adversarially coherent, environmentally portable, and verifiably assured.

URDev4ever /LATT

LLM Attack Testing Toolkit is a structured methodology and mindset framework for testing Large Language Model (LLM) applications against logic abuse, prompt injection, jailbreaks, and workflow manipulation.

scthornton /Chain-of-Thought-Reasoning-Attacks

Breaking Chain-of-Thought: A Comprehensive Taxonomy of Reasoning Vulnerabilities in Production AI Systems

DUBSOpenHub /havoc-hackathon

Pit AI models against each other. Score them sealed. Crown a winner. All built using the GitHub Copilot CLI. ⚡

Alessandro624 /BandAI

BandAI is a CrewAI-powered multi-agent orchestration platform that automates the discovery, adversarial legal evaluation, and proposal generation for Italian public tenders (bandi)

saad2134 /logic-shield

🛡️ An AI-powered debate training and communication risk analysis platform. LogicShield strengthens your arguments, detects logical fallacies, and evaluates reputational risk before you publish, pitch, or perform.

nabielnovelkhubran /Should-AI-Buy

2nd Place Winner @ Alpaca x Lablab.ai Hackathon. Autonomous multi-agent trading council with adversarial thesis falsification, deterministic risk gates (25% max allocation), live paper execution, and thesis invalidation monitoring on Alpaca Markets.

jasoncobra3 /LLM_Sentinel

LLM Sentinel Red Teaming Platform is an enterprise-grade framework for automated security testing of Large Language Models, detecting vulnerabilities such as jailbreaks, prompt injection, and system prompt leakage across multiple providers, with structured attack orchestration, risk scoring, and security reporting to harden models before production

purplesectools /AIvsAI

Adaptive red teaming for LLM applications using iterative AI-vs-AI attack orchestration.