AI guardrails & agent security
3,625 agents · most starred first
AI guardrails & agent security, page 11 of 37
AI Security Research Radar — AI for security and security for AI
The open standard for AI agent integrity. Evaluate, enforce, and prove that autonomous agents are adversarially coherent, environmentally portable, and verifiably assured.
Security boundary CLI proxy between AI coding agents and your shell — redacts secrets, neutralizes prompt injection, wraps untrusted terminal output, and audits every command. Single Rust binary.
The open standard for telling AI not to hallucinate.
Open source control-plane proxy for AI agents: identity, policy, budget, audit for MCP and beyond
A visual, source-grounded guide from agent architecture to secure agentic AI systems.
Five-level game where players try to talk a chatbot into leaking a password, each level guarded harder than the last. Flask API behind it.
Policy-Gated Governance for Operational AI Agents
Guardrails for LLM answers. GroundLens checks an LLM answer against its sources without a model, decides PASS, REVIEW or FAIL under your policy, and keeps a signed record.
Zero-trust security and governance framework for autonomous multi-agent AI networks. Implements the CSA Agentic Trust Framework to mitigate OWASP Top 10 Agentic Risks using SPIFFE/SPIRE identities, OPA/Rego policy-as-code, and semantic guardrail gateways.
An offensive security fuzzer for LLM agents that systematically tests prompt injection across user input, indirect content, memory, and tool definitions, with a focus on whether an injection causes unauthorized agent actions.
Open Guard Layer (OGL-Mini): a full-featured, fast and lightweight security model for AI agents with ready-to-use TypeScript / Python / Go modules. Use this layer to protect your prompts in AI agents and applications.
GenPark AI Agent Skill - Risk-coverage trade-off optimization; controls agent abstention, fallback escalation, and human-in-the-loop triggers under dynamic risk budgets.
Autonomous Agent Double Ratchet Key Derivation & Forward-Secret End-to-End Secure Channel Engine
Autonomous Agent Shamir (k, n) Secret Sharing Threshold Scheme with Polynomial Interpolation
Autonomous Agent Verifiable Delay Function (VDF) Sequential Squaring & Rapid Proof Verification Engine
Autonomous Agent Non-Interactive Schnorr Zero-Knowledge Proof & Fiat-Shamir Heuristic Engine
Weekend lab for the OWASP Top 10 for Agentic Applications. Build an agent, break it five ways, then harden it.
Multi-tenant security isolation guard enforcing memory boundary compartmentalization
step-labeled benchmark of injection-hijacked LLM agent trajectories (12,536 trajectories, 71,024 labeled steps)
gradio region:us
The safety layer for the AGI era. Runtime guardrails for autonomous AI agents.
5,660 AI security links: prompt injection, LLM & AI agent security, MCP, coding agents, agent skills & supply chain, red teaming, guardrails, sandboxing, incidents & CVEs, governance. 570 arXiv papers, 1,155 GitHub repos, 570 X posts, 29 categories of plain markdown you can grep, diff and fork. Obsidian vault export. Added, not endorsed.
Local Hybrid System-One Decision Engine (JevCompat 32/32 MUST, AnyJev L0 Prefill, StartLux Cascade, OpenJev-RLCD² Trainer, Zero-Shot AI Safety Guardrails)
Protect sensitive tool output and cross-tool data flows in DeepSeek Harness, with deterministic rules, provenance receipts and a bilingual live security beacon.
Deterministic Python Model Context Protocol (MCP) server & autonomous agent skill for Openclaw Zk Origin.
Multi-vector heuristic prompt injection and jailbreak detector analyzing system overrides, delimiter breakouts, and base64 payload evasion.
🛡️ Safe AI Agents through Action Classifier
Complete alignment and safety architecture for autonomous AI agents. Membrane security, continuous trust verification, epistemic hygiene, emergence protocols, and defence tools. 30 years of cybernetic research — deployable today.
Runtime Reliability Infrastructure for LLM Pipelines — enforce deterministic rules, measure the Balance Index, and audit every decision.
Control Plane Integrity Tool for AI Agents. Cryptographically seal, verify, and audit the markdown files that control your AI Agents.
Security & Compliance bodyguard for OpenClaw agents
Auth for AI agents and humans. First-class agent identity, MCP OAuth 2.1, delegation, audit. TypeScript, edge-native, MIT.
Supply chain security runtime for AI coding agents
Cycles budget and action guard for OpenClaw agents
Policy engine for AI agents — enforceable rules, risk limits, approval gates, obligation tracking, and violation detection. One .acon file. Rust core + MCP server.
AgentGovernor is a declarative AI workflow planner that enables secure, auditable, and cost-efficient MCP tool execution through validated PlanYAML workflows.
URLCheck MCP security scanner plugin for OpenClaw. Analyzes URLs for phishing, malware, and suspicious patterns before agent navigation.
Detect if content is generated by AI
SwarmShield is an enterprise-grade security system for swarm-based multi-agent communications, providing military-grade encryption, secure conversation management, and comprehensive audit capabilities.
What can your AI agent do to the real world? Scan your code. See which tool calls have zero checks
Open-source security scanner for AI agent skills
Local-first security scanner for AI Skill bundles. 72 rules, < 2s, zero LLM cost. Apache-2.0. SaaS at skillguard.vip
Cryptographic audit receipts for AI coding agents. Ed25519 + Merkle + RFC 3161 TSA. Supports Claude Code, Codex CLI & Cursor.
Pre-execution policy engine for AI agents. Every tool call checked before execution.
🛡️ Open-source safety guardrail for AI agent tool calls. <2ms, zero dependencies.
The Proxy-Free Safe Pause & Resume Layer for AI Agents. In-process circuit breaker. 2 lines of code. Patent pending.
GatewayStack governance layer for OpenClaw — identity, scope, rate limiting, injection detection, and audit logging for every tool call
Test lab for RAGdrag — vulnerable RAG target, sample results, and test suite
bondage-core is the local C launcher/policy engine.
Comprehensive, auto-updating literature review of GenAI & LLM security research, standards, tools, and resources. 100+ curated entries with interactive webapp.
Local encrypted vault MCP server. AES-256-GCM + Argon2id. Resolves {{PLACEHOLDER}} secrets at runtime so LLMs never see real API keys. Works with Claude Desktop, Cursor, Continue.dev, Cline, Open WebUI.
Agentic AI Security & Multi Agent Management: Agent Agnostic Platform. Offers Protection against Prompt Injection via Granular Action Scope & Cryptographically Protected Instruction Channel
AI agent & MCP trust registry — independent, evidence-based trust scores for 1.3K+ open-source AI agents and MCP servers. Runtime-trust calibrated (MCP, dependencies, provenance drift), with side-by-side comparison and embeddable live badges. Ranked by verifiable signals, not hype.
Pre-deploy static scanner for risky AI agent capabilities. Local-only. No telemetry. No network calls during scan. Redacted output. MIT.
Open, privacy-bounded assurance for AI agents: containment provenance, identity passports, authorization twins, OTel evidence, CI gates, and OSCAL.
A practitioner maturity model for securing conversational and agentic AI.
gradio region:us
gradio region:us
gradio region:us
An interactive, chat-driven learning system and production reference architecture for recommendation-system engineering with layered LLM guardrails, built with FastAPI, LangGraph, Ollama, and pgvector, fronted by a modern Next.js / React 19 App Router lab UI.
Reliability & security proxy for the Model Context Protocol (MCP). Self-healing connections, runtime tool-poisoning/shadowing defense, and NIST AI RMF + OWASP LLM Top-10 audit trails for any MCP server. Works with Claude, Cursor, Cline, Windsurf.
Deterministic governance for AI agent tool calls in Claude Code, MCP, LangGraph, or any agent loop. Pin approved capabilities, detect drift, and refuse unsafe or unapproved actions before execution. No model in the decision path.
TrustOS is an AI Permission Layer that sits between AI agents and real-world actions.
A credential is not authorisation. ACP moves the authorise/refuse decision for AI agent actions outside the model, where prompt injection cannot reach it. Specification, Dafny proofs, and a reference implementation — every claim replays locally.
peft safetensors cybersecurity compliance gdpr rgpd
Trust infrastructure for autonomous AI agents: decision assurance, action authorization, runtime controls, and verifiable lifecycle evidence.
Zero-trust governance middleware for Model Context Protocol deployments
Local proxy that keeps your API keys and secrets out of LLM requests. Works with Claude Code, Cursor, Codex and any tool.
Human-in-the-Loop policy gatekeeper enforcing risk tiers, reversible checkpoints & token approval bounds for autonomous agents.
In-process guardrails for AI agents. Block or rewrite tool, MCP, and model calls before they run: 100+ detectors, hierarchical policy scopes, fail-closed by default!
Universal Agentic Webhook Signature Verifier & Ingress Multiplexer. Authenticates multi-provider webhooks (Shopify, Stripe, GitHub, Slack, PayPal) with HMAC-SHA256 constant-time verification, timestamp replay window validation, and multiplexed agent event dispatch.
Enterprise Cross-System Identity & Token Exchange Bridge (inspired by WeChat Work, Tencent Cloud OAuth2 & WorkBuddy Delegated Execution). Safely manages on-behalf-of token exchange, scoped permission delegation, token lifetime refresh, and zero-leakage ephemeral credentials.
Zero-Trust Agentic Indirect Prompt Injection & Jailbreak Firewall Sentinel. Inspects untrusted external documents, emails, web pages, and user inputs for instruction-override attacks, system prompt leakage attempts, base64/hex payload smugg | GenPark: https://www.genpark.ai | LinkedIn: https://www.llinkedin.com/company/92955550
Enterprise Agentic RBAC/ABAC Least-Privilege Policy Synthesizer & Blast-Radius Auditor. Analyzes historical agent tool call audit logs, detects over-permissioned wildcards (*), synthesizes strict least-privilege JSON security boundary policies, and quantifies attack surface reduction.
Cryptographic Enterprise Audit Trail Merkle Tree Anchor (inspired by Enterprise Work Agent & Regulatory Governance). Hashes agent execution logs into tamper-proof Merkle trees, computes cryptographic inclusion proofs, and verifies audit integrity.
Ephemeral virtual identity and credential proxy sandbox for personal agents with scoped tokens and budget bounds
Enterprise Private Context & Role-Based Access Control (RBAC) Governance Sentinel (inspired by WeChat Work, Tencent Docs enterprise permissions). Inspects ingested context, classifies security clearance tiers, sanitizes cross-department prompts, and prevents data leaks.
Cryptographic JSON Web Token (JWT) engine with RFC 7519 HMAC-SHA256 signature verification, claims validation, and expiration enforcement.
Your AI agent says it's done. Assurance tells you what it didn't check.
Side-channel timing attack defense primitive providing constant-time bitwise and string digest comparison for agent security credentials.
Portable defensive application security and AI-agent security skill for Claude Code, Codex, Gemini, and ChatGPT. Covers OWASP, secure design, secure code review, MCP security, prompt injection, threat modeling, incident response, and hardening.
Security-first RAG API: prompt-injection gateway, output sanitizer, defence-in-depth middleware and full observability. Security is the architecture, not a checklist.
Open protocol that kedges passive cloud infra into self-protected agentic systems (OSS)
审计 agent 应用的 agent:每条发现附可重放 PoC,审计过程自身进哈希链。治理级 Runtime(stdlib-only)+ AGI 漏洞分类法 + 规则即数据引擎。An agent that audits agent applications.
A firewall between your computer and your AI agent. Sometimes, an agent can access your files or environment variables without your permission. It might even ask you to run a script that gives it access to your environment. Corral blocks all those attempts.
Secure credential management for AI agents
The dashcam and emergency brake for AI agents. A security proxy that physically blocks rogue LLM commands and generates cryptographically proven audit trails for enterprise compliance.
The most flexible control plane for AI Agent systems
A LangChain tool that enables AI agents to browse safely using deep Web Page inspection. Web access is the #1 attack surface for autonomous AI agents.
Safe Agentic Browsing. Assess target URLs for potential threats and alignment with the agent's browsing intent before navigation.
🛡️ The safety & validation layer for LLMs in Ruby. Prevents JSON errors, repairs AI outputs, and enforces schemas for Rails & beyond.
OpenClaw Multi-Agent Permission Governance - A soft-isolation governance solution for multi-Agent environments on OpenClaw single gateway. / 单网关多 Agent 权限治理方案 - 适用于 OpenClaw 单网关环境的多 Agent 权限隔离与协作治理方案。
Merkle Mountain Range audit ledger for tamper-evident logging and cryptographic accountability.
Core packages to implement an Agentic Control Plane. User-scoped AI governance and control plane for agentic systems. Modular identity, policy, rate limits, routing, and audit layers for LLM and tool calls.
Proxilion MCP Security Gateway is a self-hosted, Docker-ready security gateway that provides real-time threat detection (<50ms P95 latency) against insider threats, compromised accounts, and rogue AI agents by analyzing tool calls from assistants like Copilot and Claude Code, achieving a 75-85% detection rate against sophisticated attacks.
A Python implementation of the VETTING (Verification and Evaluation Tool for Targeting Invalid Narrative Generation) framework for LLM safety and educational applications.
Open-source AI agent control plane in Rust for runtime governance, agent identity, permissions, memory, human approvals, budget controls and auditable receipts. Bring existing agents or assemble one here. Active development; production coverage is being validated.
Runtime Gateway for AI Agents. 17-phase security pipeline between agents and LLM APIs. Works with OpenClaw, Paperclip, Claude Code, Cursor. Open source, MIT license.
NLIP’s Security Guidelines and Best Practices is the companion to the NLIP core specification, brought to you by the NLIP Security Working Group.