AI safety evaluation, red-teaming, LLM guardrails, vulnerability scanning, and compliance audit tools
Strix is an open-source AI penetration testing agent that automatically finds and fixes web app vulnerabilities, an automation aide for security teams.
754 structured cybersecurity skills for AI agents mapped to 5 frameworks: MITRE ATT&CK, NIST CSF 2.0, MITRE ATLAS, D3FEND and NIST AI RMF. Works with Claude Code, Codex CLI, Cursor, Gemini CLI and 20+ platforms.
Test and evaluate LLM prompts, agents, and RAG pipelines. Built-in red teaming and security evaluation for reliable AI applications.
Fully autonomous AI Agents system capable of performing complex penetration testing tasks using multi-agent architecture with support for multiple LLM providers.
SWE-agent takes a GitHub issue and automatically generates fixes using your LLM of choice, also applicable to cybersecurity auditing and competitive coding. NeurIPS 2024 paper.
OpenAI's framework for evaluating LLMs and LLM systems, providing an open-source registry of benchmarks and tools for systematic model assessment.
NVIDIA's SkillSpector inspects and evaluates the tool-use and function-calling skills of LLM agents against safety, correctness, and performance criteria.
An automated penetration testing agentic framework powered by large language models for security testing and vulnerability discovery.
OpenSandbox is an open-source, secure, fast, and extensible sandbox runtime for AI agents, developed by Alibaba.
E2B provides secure cloud sandboxes for AI agents, supporting code execution, file operations, and isolated compute as an execution layer for coding and automation workflows.
Portkey AI Gateway is a blazing fast AI gateway with integrated guardrails, routing to 200+ LLMs with 50+ AI guardrails through a single fast and friendly API.
HexStrike AI is an advanced MCP server that lets AI agents autonomously run 150+ cybersecurity tools for automated pentesting, vulnerability discovery, and security research.
Microsoft's open-source context-aware PII detection and de-identification SDK for text, images, and structured data, providing sensitive data protection for LLM applications and agents.
MCP server for Ghidra reverse engineering platform, enabling AI agents to autonomously perform binary analysis and vulnerability discovery.
Alias Robotics' open-source AI security research agent framework for multi-agent orchestration of cybersecurity tasks, integrating 300+ AI models, designed for red-team operations and security research.
NVIDIA's open-source LLM vulnerability scanner that automatically detects security issues in language models including safety vulnerabilities, hallucination tendencies, jailbreak risks, and prompt injection attacks.
OpenShell is the safe, private runtime for autonomous AI agents, developed by NVIDIA. Provides controlled execution environments and resource management.
Secure, local, cross-platform and programmable sandboxes for AI agents. Provides strict resource isolation using microVM technology.
Guardrails AI adds programmable guardrails to large language models, ensuring reliability and safety through input/output validation, structured data extraction, and custom validators.
NVIDIA NeMo Guardrails is an open-source toolkit for adding programmable guardrails to LLM-based conversational systems, supporting topic control, safety enforcement, and dialog guidance.
Superagent protects AI applications against prompt injections, data leaks, and harmful outputs, embedding safety directly into your app.
CyberStrikeAI is the system of action for AI-native cybersecurity, turning intent into governed execution chains.
Microsoft's AI Agent Governance Toolkit providing policy enforcement, zero-trust identity, execution sandboxing, and reliability engineering for autonomous AI agents. Covers 10/10 OWASP Agentic Top 10.
Tencent's full-stack AI red teaming platform integrating OpenClaw security scanning, agent scanning, skills scanning, MCP scanning, AI infrastructure scanning, and LLM jailbreak evaluation.
(24 / 67)
Most teams evaluate agents by checking a few examples. Real evaluation needs layered metrics, non-rotting datasets, and judges that push back. This article provides runnable code patterns and a practical decision framework.
Why do LLM agents hallucinate? This article traces root causes and systematically reviews practical mitigation patterns: retrieval augmentation, confidence scoring, multi-agent cross-validation, forced citation backtracking, and observability with UpTrain, Giskard, RagaAI Catalyst, Comet Opik, and NVIDIA Garak.
Based on OWASP LLM Top 10 engineering practice, this article systematically explains the seven layers of defense-in-depth for agent prompt injection: input sanitization, instruction isolation, least-privilege, output auditing, guardrails frameworks, continuous red-teaming, and kill switches -- with actionable code and toolchains.
A side-by-side comparison of five sandbox technologies, weighing latency, security, and ops cost.
Five-layer defense plus red-team loop, built on five open-source projects you can copy.
A systematic walkthrough of three major attack surfaces in AI agents, with practical code examples for prompt injection defense, tool permission scoping, and output filtering.