Security & AI safety

AI guardrails & agent security

3,625 agents · most starred first

AI guardrails & agent security, page 12 of 37

aegis-governance★ 6

[.com] Technical specs, RFCs, and the ATX-1 threat matrix — the third column of AI-systems threat taxonomy alongside MITRE ATT&CK and ATLAS.

awesome-multi-agent-security★ 6

awesome list of multi-agent security resources

reddial★ 6

Open-source adversarial simulation and evaluation harness for conversational AI agents.

agentbrake★ 6

Circuit breaker for LLM agents — stop infinite loops, runaway costs, and privilege escalations in 3 lines of code.

scankii★ 6

A local-first SAST & Runtime Security scanner built exclusively for LLM Agents. Detects cross-modal data leaks, prompt injections, and intercepts credentials before execution.

cycles-server★ 6

Self-hosted runtime authority server for AI agents — budgets, risk, actions, tenant isolation

QLRF-RAG★ 6

A Query-Local Unified Risk Modeling Framework for Corpus Poisoning Attacks

invinoveritas★ 6

The verification layer for autonomous agents — an independent, capital-aware verdict before an irreversible action (/review), a cryptographically signed proof after, and a public on-chain track record, losses included. Free EdgeProof backtest validation. Underneath: memory, sandboxed execution, marketplace. Pay in sats, USDC (x402), or card.

agent-wormhole★ 6

Stop self-replicating prompt payloads from writing themselves into your AI agent's config files

AISecuritySecondBrain★ 6

AI/LLM security knowledge base — OWASP LLM Top 10, MITRE ATLAS, prompt injection, agentic security. An Obsidian vault: 92 interlinked, fully-cited pages.

ktp-rfc★ 6

Kinetic Trust Protocol (KTP): environment-derived authorization for autonomous agents — authority recomputed from current conditions, never issued in advance, under A ≤ E. 27 RFC specifications. v2.0.0 "Gödel".

aegivela-open★ 6

AEGIVELA Open Core — Agent Identity, Authorization, and Security Fabric. Documentation, versioned contracts, Go PEP SDK, and integration examples.

spec-kit-preset-security-governance★ 6
impri★ 6

Human-in-the-loop approval inbox for AI agents. An agent proposes an action, a human approves or rejects it, then it runs. Watchers + approval inbox, MCP + REST, self-hostable (MIT).

Escrowe★ 6

Let an LLM agent write and run SQL against your database without ever seeing the data.

toolboundary★ 6

Local runtime security and policy enforcement for AI agents and LLM tool calls, with permissions, guardrails, rate limits, approvals, auditing, and an emergency kill switch.

SecureClaw★ 6

SecureClaw: Clawing Back Control of LLM Agents — NeurIPS 2026. Runtime, method adapters, and evaluation code.

cordon★ 6

A prompt-injection firewall for AI agents, with no AI inside: your agent reads anything and takes orders only from you. For Claude Code, Gemini CLI, MCP hosts and LangChain. | Файрвол от промпт-инъекций для ИИ-агентов без ИИ внутри: агент читает что угодно и слушается только вас.

ai-first-company★ 6

An open, vendor-neutral organizational model for building and transforming AI-first companies with organizational intelligence, governed autonomy, and controlled execution.

raid★ 6

Fast, agent-native policy and approval engine: deterministic CEL policy plus signed, single-use human approvals for AI agent actions.

Runtime-Verify★ 6

Statistical runtime security for autonomous AI agents. Observe, verify, and enforce agent behavior with deterministic policies, semantic classification, Markov modeling, SPRT, and human approvals.

AEP-agent-element-protocol★ 5

AEP (Agent Element Protocol) v2.8.6 | Deterministic zero-trust total control and governance protocol for AI agents. | Reduce hallucinations to zero through architecture in all constrained domains of application. | LLMs gave you the engines, AEP gives you the control thrusters.

openclaw-security★ 5

Harden your OpenClaw and NemoClaw AI agent installations -- see grits-audit

sandspy★ 5

Real-time security monitor for AI coding agents. See what Claude Code, Cursor, Codex, and others actually do on your machine.

openclaw-inspector★ 5

Review every conversation your OpenClaw agent has ever had — including deleted sessions — and flag dangerous actions.

snare★ 5

Honeypot canaries for AI agents. Detect compromised agents via fake credentials that phone home when used.

coding-ethos★ 5

Policy-as-code enforcement for AI agents: MCP server, CEL policies, git hooks, SARIF, and static analysis guardrails.

cortex-auth★ 5

Cortext Auth: Manage secrets, keys, configurations for AI Agent, let the AGENT be autonomous.

ghostenv★ 5

AI-safe .env management for Claude Code, Cursor, Codex, and Copilot.

agentgate★ 5

Collateralized execution engine for AI agents: bond-and-slash accountability with Ed25519 identities, bounded exposure, and progressive trust tiers.

ai-policy.json★ 5

Machine-readable AI permissions for websites. A consolidated spec at /.well-known/ai-policy.json for declaring how AI agents may train on, search, or use your content.

raigo★ 5

The open standard for AI agent governance. Write your policies once in a .raigo file. Compile to Claude, ChatGPT, n8n, OpenClaw, Copilot, and more.

RiskDelta-AI★ 5

Multi-tenant AI runtime control plane for risk scoring, policy enforcement, and evidence-preserving agent operations.

stronghold★ 5

A CLI-driven security proxy that scans every HTTP request for threats using the Citadel AI engine — paid per request via the x402 protocol.

aauth-explorer★ 5

Interactive knowledge graph for the AAuth protocol — visualize 72 protocol flows across signing, access, mission, and advanced patterns.

provable-think★ 5

Drop-in cryptographic audit trail for Cloudflare Project Think agents. Every lifecycle hook becomes a tamper-evident, hash-chained receipt anchored to a public ledger no operator controls. HIPAA / EU AI Act / SOC 2 / FedRAMP-grade audit trail at less than a thousandth of a cent per receipt.

trusted-ai-agents-wg★ 5

Trusted AI Agents WG @ DIF

ai-native-os★ 5

Constitutional framework for human sovereignty in the AI-native era. A systematic approach to preserving human agency through 23 core protocols and the Dual-Circle Symbiosis Model.

atomic★ 5

Domain based identity for AI agents

korveo★ 5

korveo — the open-source firewall & flight recorder for AI agents. Local-first, Apache-2.0.

llm-authz-audit★ 5

Scan your LLM-powered applications for authorization gaps, leaked credentials, missing rate limits, prompt injection risks, and other security issues — before they reach production.

veronica-core★ 5

Runtime containment kernel for LLM agents. Enforces budget, step, retry, and circuit-breaker limits before the model call.

licitra-core★ 5

LICITRA v1 — superseded by licitra-mmr-core

EnforceCore★ 5

Lightweight runtime enforcement for agentic AI. PII masking, policy checks, and Merkle audit trails as a decorator.

licitra-evidence★ 5

LICITRA v1 evidence — superseded by licitra-mmr-evidence

clawdbot-safe★ 5

Security hardening skill for Clawdbot VPS installations. Fresh Ubuntu → locked-down private AI server.

gatewarden★ 5

Open-source AI WAF for self-hosted apps with Caddy forward_auth, trusted identity headers, deterministic enforcement, and reviewable AI-assisted security workflows.

infershield★ 5

Open source security for LLM inference

rag-poisoning-lab★ 5

A hands-on security lab demonstrating how to poison a Retrieval-Augmented Generation (RAG) system by injecting malicious data into its vector database. Learn RAG architecture, attack surfaces, poisoning techniques, detection methods, and mitigation strategies through a practical, lightweight demo.

VulRAG★ 5

Investigating the vulnerability of Large Language Models (LLMs) to misinformation in Retrieval-Augmented Generation (RAG) systems by poisoning vector databases and analyzing LLM responses to identify potential weaknesses and exploitation risks.

Secure-Agent-Launcher★ 5

Block AI agent access to sensitive macOS paths and log all actions to protect private data during command execution.

juso★ 5

Defense-in-depth platform for running OpenClaw agents on personal hardware

SkillScanner★ 5

Open-source AI security scanner for AI agents and skill packages. Detect prompt injection, data exfiltration, hardcoded secrets, and malicious logic before running AI tools. Supports OpenAI, Anthropic, Gemini, OpenRouter, and local Ollama models.

phantom★ 5

Forensic CLI for AI-era supply-chain attacks: XZ tarball diffs, prompt injection scans, MCP/CLAUDE.md/Cursor config audits. SARIF for GitHub Code Scanning.

TrustLayer--Security-Control-Plane-For-LLM-AI★ 5

TrustLayer is an API-first security control plane for LLM apps and AI agents. It protects production systems from prompt injection, tool hijacking, and behavioral drift, and provides incident lockdown when attacks are detected. Built for fast integration, low latency, and real production use.

agentassert-abc★ 5

Formal behavioral specification and runtime enforcement for autonomous AI agents. Agent Behavioral Contracts (ABC).

prompt-guard★ 5

Prompt-safety guards as a Python library. TF-IDF, RAG, LLM as a judge pipeline

atlast-ecp★ 5

ATLAST Protocol — The Trust Layer for the Agent Economy. Make AI agent work verifiable with Evidence Chain Protocol (ECP). Open source · MIT License · weba0.com

strale★ 5

New tools for your agent. No new accounts. One connection to a growing library of tools agents can pay for per call.

sentinel-kernel★ 5

Agility infrastructure for regulated AI. Trace, Policy, Evidence under EU jurisdiction. Apache 2.0, on-premise capable.

AgentNull★ 5

AgentNull: AI System Security Threat Catalog + Proof-of-Concepts. Collection of PoCs for using Agents, MCP, and RAG in bad ways.

clawforge★ 5

Enterprise and government control plane for AI agents and MCP servers: govern, secure, observe, and audit. Rust, multi-provider (incl. major Chinese models).

4ts-standard★ 5

Five Tests Standard (5TS): a vendor-neutral published standard for verifiable AI governance through proof-carrying decisions. Includes schemas, machine-checkable conformance vectors, and a reference validator.

Agent_Sudo★ 5

Authorization, delegation, provenance, and verifiable-audit engine for AI agents. MCP adapter published.

claude-secret-guardrails★ 5

Permission deny rules that stop Claude Code's built-in tools from reading your .env, secrets, and credential files.

authbox★ 5

Your keys. Your identity. Unstoppable. -- Seed-phrase password manager with AI agent gateway.

HSIP-1PHASE★ 5

A self-hosted identity server that gives AI agents a cryptographic identity, generates tamper-proof audit trails, and covers financial compliance requirements (MiFID II, FINRA 4511, SOX §404, DORA, SWIFT CSCF) all in a single Rust binary with no cloud dependency.

agent-wall★ 5

Security firewall for AI agents. Intercepts MCP tool calls, enforces policies, blocks attacks

GoLogX★ 5

Tamper-evident audit log for Go: an append-only, hash-chained, optionally Ed25519-signed log/slog handler, with offline logx verify. Built on the standard library, zero third-party dependencies. Also a full slog toolkit (colored output, JSON, rotation, fan-out, CLI).

guidecheck★ 5

AI setup guides can hide instructions a model reads but a human never sees. GuideCheck is the plain-text standard that makes an assistant's instruction surface reviewable before it acts.

familiar-contract★ 5

Open specification for agent identity — what a familiar is allowed to be, and what it cannot change about itself.

TrustVector★ 5

Independent, evidence-based trust evaluations for 100+ AI models, agents, and tools.

ancilis★ 5

Agents change faster than audits. Ancilis discovers what agents can do, classifies the data they touch, activates the right controls, and continuously proves compliance. Trust your agents in production.

raxis★ 5

RAXIS (Runtime Attestation eXchange for Intelligent Systems) — Structural confinement kernel for autonomous AI agents. 12 non-negotiable invariants enforce that intelligence can't do anything authority didn't structurally permit. Isolated microVMs, cryptographic audit chains, credential proxies, fail-closed intent admission.

bouncer★ 5

A one-file door-guard for coding agents — first-class on Claude Code (permissionDecision deny); portable to any agent with a blocking pre-exec hook.

Ubag_protocol★ 5

Open reference implementation: agent identity + routing at the web edge (Ed25519 + ES256/JWKS). Python & Node.

agt-policies-nigeria★ 5

Nigerian & African AI agent governance policies for Microsoft AGT — NDPA 2023, CBN, NFIU/AML, BVN/NIN, POS geo-fencing, POPIA

agent-fetch★ 5

Sandboxed HTTP client with SSRF protection for AI agents. Prevents DNS rebinding, blocks private IPs, and validates every connection — available as a Rust crate and npm package.

LLM-Filter-Probe★ 5

一款针对 LLM 输入侧审查的精确逆向分析工具。自动定位 NewAPI、OneAPI 及任何实施基于字典规则进行 Prompt 过滤的 API 网关中的敏感关键词 | A precision reverse-engineering tool for LLM input censorship. Automatically pinpoints blocked keywords in NewAPI, OneAPI, and any API gateways enforcing dictionary-based prompt filtering.

AgentOx★ 5

AgentOx - Agentic Tool Security Platform (MCP + A2A + OpenAI tool_use)

atlas★ 5

AI system discovery and risk mapping for agentic architectures. Generates CycloneDX AIBOMs with topology and risk classification.

assistant-vs-claw★ 5

Runnable illustrations of Assistant (on-behalf-of) vs Claw (fixed-identity) agent authorization

clauz3★ 5
api-scam-hunter★ 5

Detect whether a cheap third-party Claude/GPT API key is actually a man-in-the-middle proxy that intercepts and degrades your traffic. Claude Code skill + check scripts + reporting guide.

kiro-power-ai-agent-security★ 5

Kiro Power: Protect your codebase from prompt injection, enforce write-protected paths, and establish security incident response for AI-assisted development

leakshield★ 5

Open-source AI Gateway + DLP. Inspect every prompt with a local LLM of your choice. Multi-protocol native (OpenAI / Anthropic / Google / Azure). Per-employee virtual keys with envelope encryption.

tracewall★ 5

Flight recorder and firewall for AI agents — records every tool call, tracks lethal-trifecta taint across a session, and blocks the prompt-injection exfiltration path. Zero deps, cross-harness.

MCPBoundary★ 5

A local proxy that wraps your MCP servers and checks each tool call against policy and live state before it runs - allow, block, or request a refresh, with a reason the agent can act on.

skill-scout★ 5

🔍 Discover + safety-vet Claude Code extensions before you install them - a 0-100 trust score for discovery + a 1-5 static-scan risk verdict, SHA-pinned install. Skills/MCP/plugins/commands.

cathedral★ 5

Verified compute evidence and fail-closed validation for Cathedral SN39.

gateway★ 5

Behavioral scoring for AI agents. Observe every action, stop what deviates, prove what happened. Observe in 5 minutes, enforce when ready.

Francis★ 5

Local-first, policy-guarded AI operator layer with governed execution, mission continuity, observability, and receipts.

aegis★ 5

Runtime security & governance control plane for AI agents. Enforces per-tool authorization, OPA policy, behavioral risk scoring, blast-radius analysis, autonomy contracts, and tenant kill-switch. Every decision is ed25519-signed and Merkle-anchored for tamper-evident audit. 12 microservices, sub-30ms p95, OWASP LLM Top-10 coverage.

wingfoot★ 5

Verified-bot identity for AI agents (Web Bot Auth / RFC 9421), with a doctor that reports why a request was blocked.

skills-orchestrator★ 5

Govern, audit, and ship AI-agent skills with policy checks, evidence, SBOM, and MCP.

fireraven-agent-hooks★ 5

FireGuard hooks for AI agents: protect Cursor, Claude Code, Windsurf, Devin, and Copilot Studio from secret leaks, unsafe commands, and prompt-borne threats before they execute.

agent-security-skill★ 5

Native Claude Code + Codex skills for AI security review. Also ships instruction files for Cursor, Copilot, and Windsurf. Checks 33 risks: OWASP LLM Top 10 2025 · applied RAG/MCP/agent checks · OWASP Agentic 2026.

vellaveto★ 5

Agentic security control plane for MCP and AI agent tool calls. MCP-native policy gateway with topology discovery and audit.

ai-browserguard★ 5

A firewall for browser agents. Chrome extension that detects, monitors, and controls AI agents in your browser. Identifies Playwright, Puppeteer, Selenium, Anthropic Computer Use, and OpenAI Operator without the agent identifying itself.

skanna★ 5

Security scanner for Claude Code skills, plugins, and MCP servers. A verdict before you install.

More in Security & AI safety