AI OSINT

June 18, 2026 · View on GitHub

AI-specific only. Generic OSINT tools (Shodan, Censys, etc.) → cloud_osint. Every link verified April 2026.

LLM Vulnerability Scanners

Tool	Description	By	Link
Garak	"nmap for LLMs" — probes for hallucination, data leakage, prompt injection, jailbreaks, toxicity. 4.7k+ ⭐	NVIDIA	GitHub
promptfoo	LLM pentesting CLI. 133+ attack plugins, OWASP/MITRE mapping, CI/CD integration. Now part of OpenAI. 13k+ ⭐	OpenAI	GitHub
DeepTeam	50+ vulnerabilities, 20+ attack methods. Maps to OWASP Top 10, MITRE, NIST. Built on DeepEval.	Confident AI	GitHub

AI Red Teaming Frameworks

Tool	Description	By	Link
PyRIT	Python Risk Identification Tool. Multi-turn attacks (Crescendo, TAP). Battle-tested in 100+ Microsoft red team ops.	Microsoft	GitHub
AI Red Teaming Labs	13+ hands-on challenge labs with PyRIT integration.	Microsoft	GitHub

AI API Key Detection

Tool	Description	Link
API Radar	Real-time dashboard monitoring GitHub for leaked AI keys (OpenAI, Anthropic, Gemini). 9,200+ repos scanned.	apiradar.live
KeyLeak Detector	Web scanner with 200+ patterns including 15+ AI/LLM providers. Headless browser + network interception.	GitHub
KeySentry	Scans GitHub repos for 25+ API key formats including AI providers. CLI + web.	GitHub
KeyHacks	Validation commands to verify if discovered API keys are live/active.	GitHub

AI Chat & MCP OSINT

Tool	Description	Link
promptmap	ChatGPT Google dorks and prompt injection testing. Source of `site:chatgpt.com/share` dorks.	GitHub
Vulnerable MCP Project	Comprehensive MCP vulnerability database with CVEs, CVSS, PoCs, remediation.	vulnerablemcp.info
MCP Landscape	Academic MCP security research — attack surfaces, PoC servers, lifecycle threats.	GitHub
OSINT MCP Server	37 OSINT tools across 12 data sources via MCP for AI-assisted recon.	Glama

AI Security Knowledge Bases

Resource	Description	Link
AI Red Teaming Guide	OWASP Gen AI Red Teaming Guide with tool catalogs and MITRE ATLAS mapping.	GitHub
LLM Security Guide	OWASP LLM Top 10 2025 + Agentic Top 10 2026. Updated Feb 2026.	GitHub
AI Red Team Handbook	46-chapter curriculum with automated test runner.	GitHub
AI Pentesting Toolkit	Attack catalogs for ChatGPT, Claude, LLaMA.	GitHub

Standards & Databases

Resource	Link
OWASP Top 10 for LLM Applications 2025	owasp.org
OWASP Top 10 for Agentic Applications 2026	owasp.org
MITRE ATLAS	atlas.mitre.org
MCP Security Best Practices	modelcontextprotocol.io
NIST AI Risk Management Framework	nist.gov

🆕 v1.2.0 Tool Additions (April 2026)

MCPSafetyScanner

Field	Details
URL	github.com/johnhalloran321/mcpSafetyScanner
Purpose	Audit MCP server configurations for security vulnerabilities
By	Academic Research (Leidos)
AI-Specific	✅ Purpose-built for MCP security auditing

Tool from the "MCP Safety Audit" paper (arXiv). Evaluates MCP server definitions for prompt injection vectors, tool poisoning, and excessive permissions. Integrates with the PROTOAMP framework for bridging agent security benchmarks to MCP infrastructure.

Cisco AI Supply Chain Scanners

Field	Details
URL	Cisco Talos GitHub (open source)
Purpose	Security scanning for MCP, A2A, agentic skill files, pickle files
By	Cisco AI Threat Intelligence & Security Research
AI-Specific	✅ Purpose-built for AI supply chain security

Released with Cisco State of AI Security 2026 report:

Structure-aware pickle fuzzer — adversarial pickle file generation
MCP server scanner — MCP implementation vulnerability detection
A2A (Agent-to-Agent) scanner — Google A2A protocol security
Agentic skill file scanner — malicious skill definition detection

DorkEye

Field	Details
URL	GitHub (search "DorkEye OSINT framework")
Purpose	Automated Google Dorking with multi-agent analysis pipeline
By	Open Source Community
AI-Specific	⚡ Configurable for AI infrastructure dorking

Python OSINT framework: interactive wizard, dork generator, multi-agent analysis (SQLi detection, secrets scanning, security analysis), HTML report export. Can be configured with AI-specific dork sets from this repository.

Claude Code Security

Field	Details
URL	claude.com/solutions/claude-code-security
Purpose	AI-powered static analysis — finds context-dependent vulnerabilities that rule-based tools miss
By	Anthropic
AI-Specific	✅ AI model performing security analysis

Limited research preview for Enterprise/Team. Free for OSS maintainers. Built on Claude Opus 4.6 (500+ vulns found in production OSS). Reads code like a human researcher — traces data flows, catches business logic flaws.

⚠️ Defensive capability that Red Team operators should be aware of.

ATTESTMCP

Field	Details
URL	Academic (arXiv: "Breaking the Protocol")
Purpose	MCP protocol extension: capability attestation + message authentication
By	Academic Research
AI-Specific	✅ MCP protocol hardening

Reduces MCP attack success from 52.8% → 12.4% with 8.3ms overhead. Adds capability attestation, message authentication, and trust propagation controls. Backward-compatible.

Vulnerable MCP Database (UPDATE)

Field	Details
URL	vulnerablemcp.info
Status	UPDATED: 50 vulnerabilities, 13 critical, 32 researchers

New entries: CVE-2026-25536 (TypeScript SDK), CVE-2026-30615/30624/30616/40933 (Ox batch), CVE-2026-33032 (nginx-ui MCPwn), CVE-2026-27825/27826 (Atlassian). Academic analysis of 67,057 MCP servers across 6 registries.

🆕 v1.4.0 Tool Additions (June 2026)

All AI-specific, open-source, and verified against their repos (June 2026). Theme: MCP / agent-skill auditing — the defensive response to the OpenClaw crisis.

Agent Threat Rules (ATR)

Field	Details
URL	github.com/Agent-Threat-Rule/agent-threat-rules
What	Open detection standard — "like Sigma, but for AI agents." 425 rules covering skill compromise, MCP abuse, prompt injection, exfiltration. Maps to OWASP Agentic Top 10, MITRE ATLAS, NIST.
By	Community / OWASP A-S-R-H; shipped in Microsoft AGT, Cisco AI Defense, MISP
AI-Specific	✅ Purpose-built detection format for agent/skill/MCP threats

Cisco DefenseClaw (governance suite)

Field	Details
URL	github.com/cisco-ai-defense/defenseclaw
What	Open-source agentic governance layer: admission control that scans skills, MCP servers, plugins, and generated code before they run. Bundles skill-scanner, mcp-scanner, a2a-scanner, CodeGuard static analysis, and an AI-BOM generator.
By	Cisco AI Defense (Apache 2.0)
AI-Specific	✅ Built for OpenClaw / agent ecosystems

Cisco MCP Scanner

Field	Details
URL	github.com/cisco-ai-defense/mcp-scanner
What	Standalone CLI / REST scanner for MCP servers using YARA + LLM-as-judge + Cisco inspect API. Static/offline mode for CI/CD and air-gapped use; pip-audit integration for vulnerable Python deps.
By	Cisco AI Defense
AI-Specific	✅ MCP-specific

Cisco Skill Scanner

Field	Details
URL	via cisco-ai-defense org
What	Detects prompt injection, exfiltration, and malicious patterns in agent skills (Cursor, Claude Code, Codex).
By	Cisco AI Defense (Apache 2.0)
AI-Specific	✅ Agent-skill-specific

AgentAuditKit

Field	Details
Install	`pip install agent-audit-kit`
What	77 rules across 13 scanners auditing 13 AI agent platforms (Claude Code, Cursor, Copilot, Windsurf, Amazon Q, Gemini CLI). Checks MCP config exposure, hardcoded keys, invisible-Unicode tool-description hijacking, taint analysis. Maps to OWASP MCP Top 10.
By	Independent (Sattyam Jain)
AI-Specific	✅

mcp-audit

Field	Details
What	Offline MCP config auditor across 8 MCP clients; cross-server attack paths, IDE extension security, SAST for Python/TypeScript (37 rules). Maps findings to OWASP MCP Top 10.
By	Independent (Adam Dudley)
AI-Specific	✅

SkillSpector

Field	Details
What	Agent-skill security scanner — 64 vulnerability patterns across 16 categories; 0–100 risk score with severity labels. Audits skills before deployment.
By	Independent
AI-Specific	✅

VIPER-MCP

Field	Details
What	Combined static + dynamic (taint-style) analysis framework for MCP servers. Ran across ~40,000 server repos → 106 zero-days / 67 CVEs.
By	Academic research
AI-Specific	✅ MCP-specific

Other defensive references worth tracking (knowledge bases, not tools): awesome-agent-skills-security, SAFE-MCP (OpenSSF), and the NSA MCP design-considerations guidance (June 2026).