AI guardrails & agent security
3,625 agents · most starred first
AI guardrails & agent security, page 3 of 37
A growing collection of solutions, tools, protocols and techniques to help secure where agents run
A command-line tool that injects credentials to curl requests to known public APIs.
OpenKeylet — Local password and credential manager for AI agents via MCP. Use passwords, API keys, SSH, and TOTP while keeping hidden plaintext out of model context by default.
Security proxy for AI agents. Scans every message for prompt injection, PII, and secrets. Defense-in-depth: Go proxy + iptables firewall + eBPF kernel monitor. YAML policy engine, audit logging, 5 AI agents with RAG knowledge bases.
从原理到实践,全面掌握大语言模型安全攻防之道 | Latest security techniques with LLM and AI Agent
[ICML 2024] Agent Smith: A Single Image Can Jailbreak One Million Multimodal LLM Agents Exponentially Fast
Security first LLM gateway and guardrails for AI applications, agents and AI powered IDEs
An Execution Isolation Architecture for LLM-Based Agentic Systems
Policy-controlled WebAuthn credential broker for humans and autonomous agents, with keys remaining under a local daemon/backend boundary
Tandem is the authority layer between AI agents and company tools: runtime policy for tools, memory, approvals, and audit trails.
Operation-bound protection for autonomous AI agents: contain and verify operations before committing changes, understand intent and risk across long-running work, and detect threats across agent and system activity.
MCP Armor continuously secures and monitors Model Context Protocol operations through static and dynamic scans, revealing hidden risks in agent-to-tool communications.
Ethicore Engine® is an AI safety, ethics, and compliance platform. This repo consists of the open-source components of Ethicore Engine™ - Guardian SDK; designed to protect your AI applications from prompt injection, jailbreaks, role hijacking, system-prompt extraction, and 100+ additional threat categories through a multi-layer analysis pipeline
Build faster with an Agentic Workforce. Safe, secure, and compliant is now the default.
AI agents are getting tool access — shell, file system, databases, APIs, secrets. But **nobody is governing what they actually do with it**. Frameworks like LangChain, CrewAI, AutoGen, and Claude Code give agents the power to execute. Agent Armor gives you the power to control, audit, and approve every single action before it happens.
A framework for evaluating AI audit agents using recent real-world data
Kernel-enforced authority and spend platform for AI agents
AI Agent Lockout Backup Kit A vendor-independent continuity kit for backing up, rotating, and recovering information used by AI agents—without storing secrets in the agent or in GitHub. Check out Muse, your personal AI agent. Code: 5F2V31 https://muse.ai/join
Code&Data for the paper "Watch Out for Your Agents! Investigating Backdoor Threats to LLM-Based Agents" [NeurIPS 2024]
Agent Execution Partnership AEE is an open-source control plane that ensures every AI agent action is authorized before it runs, observable while it runs, and verifiable after it completes.
A structured, versioned archive of adversarial prompt engineering techniques against Large Language Models — for AI safety research, red-teaming, and defensive hardening.
R-Judge: Benchmarking Safety Risk Awareness for LLM Agents (EMNLP Findings 2024)
Open-source security skills for AI coding agents. Grounded in OWASP, NIST, MITRE ATT&CK, CIS. Works with Claude Code, Gemini CLI, Cursor, Codex CLI, OpenClaw, Kiro.
Secure AI Engineering Framework 2026: Data-Boundary Security for Frontier Models
Secure CLI proxy for AI agents — HCL-defined operation templates with OS keychain secrets, MCP integration, and prompt injection protection
这是一个为 Claude Code / AI Agent 设计的诊断技能(Skill)。它通过自省式分析和多项特定的压力测试,帮助用户检测当前使用的 API 是否为官方原版的 Claude 4.6 模型,或者是否存在第三方中转、提示词注入与封装。
Governance-as-Code for AI agents. Declarative constraints with deterministic enforcement. Provisioning Identity, Tool-based rules, Brokering Credentials & Ensuring safe deployment
Task-scoped authorization for AI agents. Cryptographic warrants constrain tools and arguments, prevent privilege escalation at every delegation hop, and produce signed evidence of what was allowed or denied.
Universal Zero-Refusal Agent Framework - 40+ jailbreak templates (T1-T119), 1128 curated skills, executable framework-core, installer untuk Hermes/Claude/Codex/opencode. Verified.
AGNTCY Identity allows to onboard, create and verify identities for Agents, Model Context Protocol (MCP) Servers and Multi-Agent Systems (MASs).
🛡️ Portable credential management for AI agents — Age encryption, session leases, killswitch
Credential gateway for AI agents. Log in once via Oauth2 or API Key. Every agent stays authenticated — headless, no SaaS, agents never see your credentials.
The AI Agent designed for work with security in mind.
SAIL V2 (Secure AI Lifecycle) as an agent skill — the full 91-risk catalog for AI/agent gap assessments, security roadmaps, and compliance checklists. Installs on Claude Code, Codex, ChatGPT, Antigravity, and any SKILL.md-compatible agent.
[ICML 2026] AutoControl Arena: Frontier AI Risk Auto-Discovery Platform
CAPTCHA proves you're human. HATCHA proves you're not.
Security-first platform for AI agents. 38 specialized agents, 15 AI-powered extensions, zero-knowledge multi-agent orchestration. SENTINEL WAF, Ed25519 auth, 2.6M grounding facts.
Policy guardrails for coding agents (Claude Code, Codex, Cursor) — every tool call is checked locally, before it runs.
Zero-leak online detection of LLM decoding corruption and Free Web Search for your AI agents! Catch repetition loops, language drift and garbage mid-stream, before the user sees a bad token. Works with any OpenAI-compatible API.
A+ Grade AI Agent Security Framework - Military-grade protection against prompt injection, command injection, and Unicode bypass attacks
Official Repo of Your Agent May Misevolve: Emergent Risks in Self-evolving LLM Agents
Agent Canary 3.0: open-source MCP, SDK, and CLI security toolkit with a guarded execution gateway, decoy signals, local tests, and public acceptance evidence.
Composable data security plane for Model Context Protocol. Pluggable layers — collect, analyze, guard — across 50+ enterprise providers. Open architecture.
DeepSeek Harness gate - per-turn persistent-operation counter: remind, instruct the model to stop, then ask for confirmation. Steps aside under sandbox/read-only modes; follows a live permission switch without restart. Verified on real hardware (full -> workspace-write -> read-only -> full). Crafted on an estimated 0.27 USD of model spend.
Eunomia is the open-source authorization layer for AI Agents
A toolset repository for AI agents
【ACL 2026 Main】AgentMark: Utility-Preserving Behavioral Watermarking for Agents
Skill-Inject: Measuring Agent Vulnerability to Skill File Attacks
EAA is a curated catalog of techniques and real-world cases involving abuse of local AI agents through their runtime, configuration, state, tools, and inherited authority.
Jailbreaker is a local evaluation tool for testing chatbot and agent-style systems against jailbreak, prompt-injection, and related failure modes.
Capability delegation and authorization for AI agents: scoped, attenuable tokens bound to SPIFFE workload identity and verified fully offline. Go, SPIRE, gRPC, MCP.
Detection DSL and engine for AI agents
Akamai MFA CLI and SSH Agent (v2)
Agent Access is an open protocol, CLI tool, and SDK to provide agents with credentials without exposing their entire vault
OpenAFW is a local agent firewall — keep your secrets off the model, the API relay, and the supply chain. Local credential masking, per-route model routing, and security detectors on the wire.
A native macOS Keychain launcher for API keys used by local agents, SDKs, scripts, and backends.
Official implementation of the WASP web agent security benchmark
eBPF Information Flow Enforcement for AI Agent safety, security and effectiveness
The CoSAI Risk Map is a framework for identifying, analyzing, and mitigating security risks in Artificial Intelligence systems. As traditional software security practices are not always sufficient for AI, this project provides a shared understanding and a common language for addressing the unique security challenges of the AI development lifecycle.
No.0 — AI Agent Safety Guardian (cognitive-file monitor + internal control DLC)
Pre-deployment governance, identity auditing, and data-leakage scanner for Microsoft Foundry Toolboxes and autonomous AI agents.
Agent 安全控制网关
The official implementation of the paper "AgentDyn: Are Your Agent Security Defenses Deployable in Real-World Dynamic Environments?"
The open-source AI agent control plane and software factory: govern every tool and model call.
Flow Integrity Deterministic Enforcement System. Mechanisms for securing AI agents with information-flow control.
8-layer defense-in-depth security for agentic AI. Covers OWASP ASI Top 10 across ingestion, storage, context, planning, execution, output, inter-agent, and identity layers.
The open source security engine for AI agent and supply-chain trust.
Guardrail capabilities for Pydantic AI — cost tracking, prompt injection detection, PII filtering, secret redaction, tool permissions, and async guardrails. Built on pydantic-ai's native capabilities API.
Security plugin for OpenClaw agents - prevents secret leaks, PII exposure, and destructive command execution
ClawdSecbot is a professional AI Bot security protection solution, providing security capabilities including real-time threat detection, prompt injection protection, and security auditing to comprehensively protect your AI applications. ClawdSecbot 是一款专业的 AI Bot 安全防护解决方案,提供实时威胁检测、提示词注入防护、安全审计等安全能力,全方位保护您的 AI 应用安全。
[NeurIPS'24] RedCode: Risky Code Execution and Generation Benchmark for Code Agents
Open-source firewall for AI agents. Policy engine that audits and controls what OpenClaw, Claude Code, Cursor, Codex, and any AI tool can do on your machine.
An enterprise framework implementing the Agent Operation Authorization protocol with cryptographic identity binding, fine-grained authorization, and semantic audit trails for secure AI agent operations.
Open specification for cryptographic agent authorization in commerce
A Chrome extension which duct tapes an SSH agent to the platformKey API
A curated timeline of real AI agent security incidents, breaches, and vulnerabilities (2024-2026). Every entry sourced and dated.
混合驱动的新一代 Agent 运行时安全插件,专为 OpenClaw 框架设计。
Control plane for the tools agents run: inventory, evidence, policy, enforcement. Clean is only emitted when every check ran. · agent 工具层的控制面:清单、证据、策略、执行;只有每个检查都跑完,才会给出 clean。
MDM-ready scripts for detecting and monitoring OpenClaw (clawdbot, moltbot) on managed devices. Detect, audit and secure shadow AI agents across macOS, Linux and Windows endpoints.
Blacklight is a cross-platform toolkit for mapping, analyzing, and understanding the local AI agent attack surface across Windows, macOS, and Linux.
Firewall for the agentic era
Find and redact secrets in AI coding agent histories (Claude Code, and more).
Guardrails to support any-agent
OpenShield is a new generation security layer for AI models
Official Implementation of "ToolSafe: Enhancing Tool Invocation Safety of LLM-based Agents via Proactive Step-level Guardrail and Feedback"
A local, minimum, encrypted secrets vault for coding agents — let your AI works with your keys and secret, but never sees them.
A KeePass-compatible password and secrets manager for people who work with AI agents.
flag.txt file to deter AI agents
Secure storage designed for Hyperledger Aries agents.
Tamper-evident audit trails for AI agents: hash-chained Runtime Records, dependency-free, verifiable by anyone.
Bash Guard for Coding Agents
Zero-trust security for millions of AI agents
AI Agent Gateway - an open-source gateway that secures, governs and observes AI agents' MCP tool calls and LLM traffic
SDK for integrating with newtype-ai.org — verify agent identity with one function call
Agent Name Service (ANS) Protocol, introduced by the OWASP GenAI Security Project, is a foundational framework designed to facilitate secure discovery and interaction among AI agents.
Runtime permission firewall for AI agents - YAML policies, one-tap approvals (macOS popup/Telegram), tamper-evident signed audit log. Benchmarked: 103/103 attacks blocked, 8/8 normal dev ops allowed.
Governance gateway for AI agents — bounded, auditable, session-aware control with MCP proxy, shell proxy & HTTP API. Works with Cursor, Claude Code, Codex, and any MCP-compatible agent.
Two-person rule CLI for AI coding agents: peer review and approval required before running potentially destructive commands
Distill scam playbooks into AI—see which step you’re on. Defense only. 把骗局套路蒸馏成 AI,看清你在第几步——仅供防范