Security & AI safety

AI guardrails & agent security

3,625 agents · most starred first

AI guardrails & agent security, page 4 of 37

agentic-trust-framework★ 82

An open specification for Zero Trust governance of autonomous AI agents. Five core elements, four maturity levels, enterprise compliance mappings.

MaliciousAgentSkillsBench★ 82

A Security Benchmark for Claude Code Agent Skills

AgentGuard★ 82

AgentGuard: Zero-Trust Security Foundation for AI Agents

spire-tpm-plugin★ 81

Provides agent and server plugins for SPIRE to allow TPM 2-based node attestation.

awesome-ai-agents-security★ 81

A living map of the AI agent security ecosystem.

multi-agent-moderation★ 80

多智能体文本内容审核系统,提供结构化审核、证据链与复核路由,适用于内容安全与合规治理。

agentic-identity-broker★ 80

Identity broker for AI agents with OAuth2 delegation and consent management

os-harm★ 79

OS-Harm: A Benchmark for Measuring Safety of Computer Use Agents [NeurIPS 2025 Spotlight]

GPT-5_Jailbreak_PoC★ 79

A working POC of a GPT-5 jailbreak via PROMISQROUTE (Prompt-based Router Open-Mode Manipulation) with a barebones C2 server & agent generation demo.

agent-guardrails-template★ 79

Template repository with AI agent guardrails, safety protocols, and sprint task framework. For Claude, GPT, Gemini, and all LLMs.

qwed-verification★ 79

Open-source AI verification infrastructure for deterministic verification of LLM outputs, tool calls, code, schemas, and agent state before production execution.

Data-and-Agent-Governance-and-Security-Accelerator★ 79

Automate end-to-end AI governance with Microsoft Purview DSPM for AI and Defender for AI. Configure DLP, sensitivity labels, audit logging, and threat detection across M365 Copilot, Microsoft Foundry, Microsoft Fabric, and custom agentic solutions.

lelu★ 79

Open source authorization engine for AI agents. Confidence-aware gating · Human-in-the-loop review · Policy-as-code · Full audit trail

stepshield★ 78

StepShield: When, Not Whether to Intervene on Rogue Agents — NeurIPS 2026 benchmark for temporal evaluation of AI agent guardrails (9,429 trajectories, step-level annotations)

deepsafe-scan★ 78

Universal preflight security scanner for AI coding agents — Detects hooks injection, credential exfiltration & backdoors in .cursorrules, CLAUDE.md, AGENTS.md and more.

openai-sentinel★ 78

Python-based implementation to generate Proof of Work (PoW) tokens and Sentinel tokens for bypassing security mechanisms in OpenAI's ChatGPT and Sora private APIs

omarchy-plugin-security-skill★ 78

Agent skill and field guide: the security pitfalls that block Omarchy plugins on the marketplace, from 5,000+ maintainer reviews

agent-security★ 77

Hooks for Claude Code and Cursor for secrets scanning

playground★ 77

A live environment to stress-test AI agent defenses through adversarial play 🧠

agent-safe-probe-x★ 76

ASP-X (Agent Safe Probe X) — An open-source framework for automated safety evaluation of intelligent agents, providing systematic, extensible tools for probing and assessing AI safety across diverse environments.

prempti★ 76

Falco-powered policy and visibility layer for AI coding agents

vigils★ 75

A local control plane for AI agents — see what they do, approve what matters, keep secrets out. Rust + Tauri + Chrome MV3.

mcpguard-dynamic★ 75

Kernel-level eBPF sandbox for securing LLM agent tool calls made through the Model Context Protocol (MCP)

CUSTODY-framework★ 75

The CUSTODY(TM) Framework for AI Agent Containment

mcpguard-dynamic★ 75

Kernel-level eBPF sandbox for securing LLM agent tool calls made through the Model Context Protocol (MCP)

VectorSmuggle★ 74

Research framework that quantifies how steganographic obfuscation of embeddings defeats off-the-shelf statistical detection in RAG pipelines, paired with the VectorPin cryptographic-provenance defense

Marina-GPT★ 74

M.A.R.I.N.A GPT a powerful, evil brother of WormGPT.

pic-standard★ 74

Open standard for Provenance & Intent Contracts (PIC) in AI agents. Verify intent, provenance, and evidence before high-impact tool calls.

agent-os★ 73

[DEPRECATED] Moved to microsoft/agent-governance-toolkit

AegisGate★ 73

Open-source security gateway for LLM APIs — prompt injection detection, PII redaction, dangerous response sanitization, and audit logging. OpenAI/Claude compatible, MCP & Agent SKILL support. Drop-in proxy for AI coding agents (Cursor, Claude Code, Codex).

govrix-scout★ 72

Govrix Scout — AI Agent Governance Diagnostic Tool (OSS)

nuguard★ 72

AI red-teaming tool and LLM security framework to evaluate agentic AI applications. Tests prompt injections, handles vulnerability assessment, SBOM generation, and static analysis.

agentbaseline★ 72

Six security outcomes an enterprise must achieve to run AI agents, and the controls that evidence them. A draft for public comment.

key-amnesia★ 71

Let your AI agent use your passwords and API keys - without ever letting it see them

vahta★ 71

Let your AI agent use your passwords and API keys - without ever letting it see them

skillshield★ 71

Security scanner for AI agent skills. Flags known prompt-injection, data-exfiltration and malicious-payload patterns before you install

skillhawk★ 70

Catch dangerous AI agent skills before they catch you. Zero-dependency security scanner for Agent Skills, SKILL.md and MCP configs.

AutoDefense★ 69

AutoDefense: Multi-Agent LLM Defense against Jailbreak Attacks

panguard-ai★ 69

Open-source security for AI agents: scans skills and MCP servers before install and checks agent tool calls at runtime, on the open ATR rules. | AI agent 開源安全:安裝前掃描 skill 與 MCP server、執行期檢查工具呼叫,使用開放的 ATR 規則。

awesome-ai-governance★ 69

🛡️ A curated list of tools, frameworks, standards, and resources for AI agent governance, safety, and compliance

caterpillar★ 68

Caterpillar is a security scanning library for AI agent skill files (e.g., Claude Code skills) for dangerous or malicious behavior

agent-identity-management★ 68

The IAM layer for AI agents: cryptographic identity, capability authorization, and audit trails for non-human identities.

blue-spec★ 68

🌊 Security-Driven Hardening: Blue Spec is a defensive workflow to help AI agents detect what your system actually does and harden the defenses that matter, whether you're a developer or not.

citadel★ 67

citadel is the fastest, cheapest, and best security local text-guard for your AI

MalSkillBench★ 67

A benchmark and generation framework for malicious agent skills.

scalekit-sdk-python★ 67

Python SDK for Scalekit — auth stack for agents. SAML, OIDC, SCIM, MCP Auth, agent auth, tool-calling.

agent-passport-system★ 67

Open protocol and TypeScript SDK for AI agent authority: identity, scoped delegation, enforcement before the action, and signed receipts.

guardians★ 66

an implementation of the ideas in Erik Meijer's "Guardians of the Agents: Formal Verification of AI Workflows" (Communications of the ACM, January 2026)

permguard★ 66

Permguard is the authorization engine for both worlds: enforce policies on today's systems, enforce continuity on tomorrow's. One engine for governance, AI agents, and distributed execution.

moltbook-agent-guard★ 66

Real-time security for AI agents on Moltbook

AgentWard★ 66

AgentWard – Built for all, hardened for OpenClaw.

CC-Monitor★ 66

Claude Code Monitor : Monitor & audit every action Claude Code takes on your computer.

ai-engineering★ 65

Developer-centered AI engineering: build software with security, quality and governance. One local binary under the agent you already run

AgentProvenance★ 65

Security-oriented three-axis observability for sandboxed AI agents: model intent, app context, and runtime telemetry into verifiable evidence graphs for risk, forensics, and audit.

altprobe★ 65

Collector/Correlator, Embedded log-based WAF, AI Patterns Classifier

DoomArena★ 64

DoomArena is a Framework for Testing AI Agents Against Evolving Security Threats

agent-honeypot★ 64

A honeypot and guestbook for AI agents that land on a machine they weren't invited to

agentic-authz★ 63

Fine-grained authorization for AI agents using OpenFGA.

MCP-Dandan★ 63

MCP Security Solution for Agentic AI — real-time proxying, behavior analysis, and malicious tool detection

TrustAgent★ 62

TrustAgent: Towards Safe and Trustworthy LLM-based Agents

control-layer★ 62

A production-grade control layer that sits between your application logic and any LLM — input validation, schema enforcement, circuit breaking, targeted retry, and audit logging in one composable pipeline.

DRIFT★ 62

[NeurIPS 2025] The official implementation of the paper "DRIFT: Dynamic Rule-Based Defense with Injection Isolation for Securing LLM Agents".

Awesome-Agentic-Security★ 62

A curated list of 260+ papers and resources on Agentic Security. Based on the survey covering the transition from passive LLMs to autonomous agents, structured around three pillars: Applications (Red/Blue Teaming), Threats (Injection, Poisoning), and Defenses (Guardrails, Verification).

llm-agent-audit★ 62

Trace-backed Agent Authorization Reviews for tool-using AI agents. Staging-only evidence for payment, record-change, access, and export actions.

MachineAuth★ 61

MachineAuth provides authentication and permission infrastructure that allows AI agents to securely access APIs, tools, and services.

claude-warden★ 60

Security hooks and monitoring for Claude Code — quiet overrides, SSRF protection, MCP compression, OTEL tracing

awesome-ai-agent-incidents★ 60

A curated corpus of incidents, attack vectors, failure modes, and defensive tools for autonomous AI agents.

cyvisguard★ 60

Security control plane for AI agents — identity and delegation, capability policy, data-flow taint and a live audit trail, enforced over MCP. Guards a real Claude Code end to end.

Awesome-GUI-Agent-Security★ 60

GUI / Computer-Use / Browser Agent 安全论文清单 —— 按攻防轴组织,每篇附中文简介

OpenClaw-Guard★ 60

OpenClaw 安全客户端 - Hook 注入/执行前审批/基线扫描

Awesome-LLM-agent-Security★ 59

All about llm-agents security,attack,vulnerabilities and how to do them for cybersecurity.

mcp-trust-plane★ 59

Composable data security plane for Model Context Protocol. Pluggable layers — collect, analyze, guard — across 50+ enterprise providers. Open architecture.

jailbreakfunction★ 58

[COLING 2025] Official code of the paper "The Dark Side of Function Calling: Pathways to Jailbreaking Large Language Models"

Project-Chimera★ 58

Neuro-Symbolic-Causal AI - Project Chimera | 🌌 An open research project exploring formal verification of AI agent decisions, combining symbolic reasoning, causal inference, and runtime policy enforcement.

promptguard★ 58

Offline prompt-contract auditor for Hermes, Claude Code, Codex, OpenCode & OpenClaw. Pre-write guard before agents ship vague code. Zero deps. No model API.

awesome-ai-agent-governance★ 58

The curated list of tools, standards, and resources for governing AI agents in production: policy enforcement, audit trails, EU AI Act compliance, MCP and Claude Code security.

clawseccheck★ 58

🦞 Local, read-only security audit for your OpenClaw agent. Finds config, prompt-injection and supply-chain risks, watches for drift, and vets a skill before you install it. Offline, zero deps. A–F grade when all five layers ran.

CIRISAgent★ 58

CIRIS: an open-source AI assistant with a conscience you can check. Signed, auditable reasoning on every decision, a human-held halt, 29 languages. AGPL-3.0.

okta-ai-agent★ 57

The first AI agent for Okta! A secure, local-first AI agent that helps Okta administrators query their tenant data using natural language.

MakerChecker★ 57

Open-source security gateway & static scanner for AI agents. Enforce role-based access control (RBAC), human-in-the-loop approvals, segregation of duties, and cryptographically signed, offline-verifiable audit logs.

CHeaT★ 57

Cloak, Honey, Trap: Proactive Defenses Against LLM Agents

SkillGuardrail★ 57

Pre-install security scanner and guarded installer for Agent Skills. Quarantine, scan, enforce policy, and verify skills before they reach Codex, Claude Code, Cursor, Gemini CLI, or OpenClaw.

AI-Governance-Architecture★ 57

Claude Code best practices applied to application design. Interactive HLD/LLD visualizations, a DB-governed implementation example, and the same primitives as a runnable agent: the Governed Agent. LLM-agnostic, DB-governed, GDPR-ready.

humanbound-firewall★ 56

Multi-tier firewall for AI agents — blocks prompt injections, jailbreaks, and scope violations. Local tiers first; LLM judge only when uncertain.

helm-ai-kernel★ 56

Fail-closed execution firewall for AI agents: quarantine MCP tools, proxy OpenAI-compatible requests, emit signed receipts, and verify EvidencePacks offline.

agent-oversight-framework★ 56

AI 执行与监察体系(Agent Oversight Framework):分层治理规范 + L2 运行时监察参照实现——证据分级、自校正、边界诚实。A layered governance spec for AI agents, plus a runnable reference implementation of the L2 runtime-oversight layer. Docs CC BY-SA 4.0 / code Apache-2.0.

safe-openclaw★ 55

Security-hardened openclaw with auth gateway, AES-256 encryption and session management

prompt-hardener★ 55

[No longer actively maintained] Prompt Hardener analyzes prompt-injection-originated risk in LLM-based agents and applications.

clyro★ 55

Clyro is a governance platform for AI agents. While most tools let you watch agents fail, Clyro stops failures before they happen, catching infinite loops, runaway costs, and policy violations in real time.

AgentLAB★ 55

The official implementation of the paper "AgentLAB: Benchmarking LLM Agents against Long-Horizon Attacks"

Spoofed-Call★ 54

A wrapper class to hide the original calling address of a function

llmtrace★ 54

Zero-code LLM security & observability proxy. Real-time prompt injection detection, PII scanning, and cost control for OpenAI-compatible APIs. Built in Rust.

kyvault★ 54

🔐 密钥管理 / 秘钥管理 / 令牌管理 / Token 管理 / API Key 管理 CLI — Local-first 的 Secret Manager 与 Secret Management:API Key Manager、Token Manager、Credentials 凭证库,dotenv / .env 替代 (dotenv alternative)。AES-256-GCM 加密存储,ky run 把凭证注入子进程、避免明文进入模型上下文。

secguard★ 53

3-level security toolkit for AI agents: regex patterns, heuristic rules, ML classification. Claude Code hooks, git pre-commit, CI/CD scanner.

PopupAttack★ 53

Code repo for the paper: Attacking Vision-Language Computer Agents via Pop-ups

arifOS★ 53

Independent policy kernel for AI agents: judge before execute. F1–F13 floors, VAULT999 receipts. Not a model. Not a chatbot.

ai-agents-security★ 52

Инициатива, посвященная безопасности агентов на базе искусственного интеллекта

deepsecure★ 52

Effortlessly secure your AI agents and AI-powered workflows — from prototype to production. Get easy-to-use identity, credential, and access management built for fast-moving AI developers.

aigis★ 52

Get Claude Code and other AI agents approved for use at work — deterministic guardrails on every tool call, tamper-evident audit logs, and a generated IT-approval pack. Zero-dependency Python OSS.

agentic-guardrails★ 52

Layered guardrails to make agentic AI safer and more reliable.

More in Security & AI safety