Security & AI safety

AI guardrails & agent security

3,625 agents · most starred first

AI guardrails & agent security, page 9 of 37

aegisora.ai★ 14

The Zero-Trust Runtime Security & Governance Layer for Autonomous AI Agents.

McpVanguard★ 13

Open-source security gateway for MCP agents and tools. Inspect tool calls before execution, enforce policy, block risky operations, and emit audit-ready evidence.

serverless-chat-langchainjs-purview★ 13

About This sample showcases how Purview API can be integrated to audit and secure AI prompts and responses

SecuPrompt★ 13

Protect your AI from Prompt Injection

Agentic-IAM★ 13

Agentic Identity & Agentic Security System

zone-of-distrust★ 13

Open security architecture for autonomous AI agents - extending Zero Trust principles

StrengthX★ 13

Check Your Password is Ever Cracked & Know About Strength of Your Password & Generate Passwords Using a Specialized AI Model (StrengthX-Dildo:V1) Dynamic Intelligent Lock & Defense Operator

agent-audit-kit★ 13

Static scanner for MCP-connected AI agent pipelines. 340 rules across 14 categories, 12 compliance frameworks, OWASP Agentic 10/10 + MCP 10/10, GitHub Action, SARIF, public CVE-to-rule ledger.

styxx★ 13

The measurement layer for machine minds. Reads what a model means and whether it holds the truth; certifies every claim re-runs. meaning_diff + OATH certify + mind profiles + live grounding signal + the cognometric instruments. No torch, no LLM in the loop for the core; MIT, open at the core.

SentinelMCP★ 13

The Open-Source MCP Firewall & Security Gateway for AI Agents. Inspect, redact, and control tool calls. Proxy mode (universal) and Inline SDK mode (Go).

LLM-agent-memory-poisoning★ 13

Codebase for Hidden in memory: Sleeper memory poisoning in LLM agents

rootsign★ 13

Rootsign is an open-source tamper-evident decision and action provenance logging library for AI agents

comply54★ 13

African AI governance — runtime compliance enforcement for AI agents operating under NDPA, CBN, NIMC Act 2026, KDPA, POPIA and 15+ African regulations. LangGraph, CrewAI, and AutoGen adapters included.

kakunin-samples★ 13

Runnable examples for AI agent identity & compliance — certificate issuance, scope enforcement, and integrations for LangChain, CrewAI, Next.js & more.

Project-Koma★ 13

Lightweight, agent-friendly AI security toolkit for Node.js & TypeScript. Check if your agent has loaded the skill, Stop prompt injection, audio hallucinations, and RAG data scraping. Zero dependencies. MIT

agent-loop-guard★ 13

Protect AI agents from runaway loops, repeated tool calls, and uncontrolled execution. Provider-independent, zero-dependency, TypeScript-first.

melra★ 13

Open-source local MCP runtime that checks AI-agent tool calls, gates risky actions on approval, verifies results, and avoids blind retries after timeouts.

agent-emergency-protocol★ 13

Agent Emergency Protocol Incident Codes: a vendor-neutral runtime incident-code language for autonomous AI agents. Created by Mitthan Meena.

prompt-injection-ctf★ 13

Free prompt injection CTF: 16 LLM security missions plus a 30-mission Agentic Security Lab (OWASP Agentic Top 10), defender mode, optional live-model grading. No signup.

leashed★ 12

AI got hands. This is the leash. Policy, audit, kill switch for any AI agent with access to your accounts.

make-trust-irrelevant★ 12

plan-bound authorization architecture for governing privileged effects in untrusted computational agents.

shark★ 12

Open Source Single Binary Authentication for the Agentic Era.

tonesoul52★ 12

AI governance framework — semantic responsibility, self-auditing memory, council deliberation. Honest about what it enforces vs. merely declares. 語魂:讓 AI 對自己說過的話負責。

scanner★ 12

Open-source CLI scanner for agentic AI components such as skills, MCP servers, system prompts

gsh-framework★ 12

Open-source hunt playbooks and a policy-driven Sentinel reference framework for detecting runaway AI agents, unauthorized tool calls, behavioral drift, DNS exfiltration, and poisoned MCP tools. MITRE ATLAS mapped.

lynx★ 12

Policy-gated, durable, audited execution for AI agent tool calls. Every action gets a five-verdict policy check, a crash-safe checkpoint, and a tamper-evident audit event — a library, not a server.

AWS-Evidence-Layer★ 12

Open-source verifiable audit infrastructure. Built for lean, pre-series B, AWS-Native teams.

arc-gate★ 12

Arc Gate — LLM proxy with prompt injection detection. Bendex Geometry.

web4★ 12

Web4: an open standard for verifiable AI presence — LCT identity, T3/V3 trust tensors, MRH context, ATP/ADP. Rust + Python SDKs on crates.io/PyPI. AGPL-3.0-or-later.

Who-Let-The-Agents-Act★ 12

Who Let the Agents Act? is an interactive agentic-security lab exploring how AI agents use tools, access data, and make decisions. Compare vulnerable, prompt-only, and hardened agents to see why prompts are not security boundaries and how authorization, least privilege, data projection, output controls, etc. keep agents safe.

control★ 12

Observe AI agent runs and control consequential actions. Signed policies, exact-action approvals, AuthZEN API vetoes and evidence. MCP gateway today; connected-channel supervision and scoped reactions toward 1.0. Experimental alpha.

aegivela-open★ 12

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

guardian-agent★ 11

Security-first AI agent orchestration platform with local LLM support, tool approvals, sandboxing, prompt-injection defenses, coding workspace, automations, and personal assistant integrations.

grits-audit★ 11

Security scanner and auto-fixer for OpenClaw and NemoClaw AI agents. Scan, score, and harden in 30 seconds.

faidd★ 11

FAIDD is a sovereign security framework designed to govern, audit, and secure the interactions of AI agents within high-integrity development environments

claude-code-hardened★ 11

🛡️ Stop your AI from pushing secrets to GitHub. Security hooks, battle-tested rules, and a live validator for Claude Code. 5 hooks · 6 rules · 1 install script · 0 dependencies.

dae★ 11

DAE is a an open, vendor-neutral security specification for AI Agents that formally separates reasoning and authority

TaG★ 11

Trust and Governance framework for AI coding agents. Local-first guardrails for spending, credentials, deployment, and cross-project isolation.

openclaw-middleware-suite★ 11

Six in-process middlewares for OpenClaw: HITL approvals, prompt-injection guardrails, PII redaction, tool-call budgets, context compaction, and complexity-aware model routing. Zero telemetry, all state local.

LyricMind★ 11

LyricMind is an AI-powered cognitive firewall that analyzes song lyrics to protect against unconscious absorption of harmful content. Using customizable frameworks and LLMs, it detects risky patterns and integrates with music players to trigger protective actions when thresholds are exceeded.

Sentinel★ 11

🛡️ Zero-trust governance for AI agents. Intercept, approve, and audit LLM actions with one decorator. Fail-secure by default.

agent-integrations★ 11

IntentFrame security plugin for Hermes Agent (Nous Research) — an external policy checkpoint that gates terminal, code, file, and cron tool calls before they run on your machine.

ironscope★ 11

eBPF-based per-tool syscall attribution and enforcement for LLM agents

agent-security★ 11

Hands-on AI agent security resource: OWASP GenAI attack/fix/verify demos on a LangChain IT-helpdesk agent, deployed to the cloud.

mcp-safeguard★ 11

🛡️ Security scanner for MCP (Model Context Protocol) servers — prompt injection, credential exposure, SSRF, tool poisoning & source-level credential-handling flaws. pip install mcp-safeguard

claude-tools★ 11

The clean slate a fresh Claude Code should start from — clean context, locked credentials, guarded egress. Isolated profile, pick-and-choose.

HolyTrinity-Benchmark★ 11

Adversarial security benchmark for agent authorization: does a compromised agent's policy-violating proposal become an unauthorized external effect? 61 trials, nine families, an independent oracle, per-mechanism ablation, confidence intervals. 0 unauthorized effects in 61 attack trials (95% CI [0.0%, 5.9%]). Reproduction is partial.

sunglasses★ 11

Open source input firewall for AI agents, beta. A local scanner checks text, code, PDFs, images, QR codes, audio and video with 1,565 patterns across 117 categories and reports findings and incomplete scans. A Claude Code hook blocks secret material in tool calls and the paths and hosts your policy

PoisonRecall★ 11

Attack taxonomy, toolkit & defenses for persistent-memory poisoning of LLM agents. Author: Krishita Sanjay Choksi.

Skill-Doctor★ 11

Multi-layer security platform for AI agent skill files. Detect malicious prompt injections, obfuscation, and vulnerable skills before they reach your agent runtime.

kavach-rs★ 11

Kavach: fail-closed runtime governance kernel for autonomous AI coding agents. Lifecycle gates, mistake ledger, 3-witness verification. Rust.

agent-latch★ 11

Local-first AI agent security scanner and MCP traffic monitor. Finds prompt injection, over-privileged tools, and secrets in agent code, manifests, and prompts; records MCP tool calls with secrets masked. CLI, pre-commit, GitHub Action, SARIF, OWASP Agentic Top 10.

aka-claude-tools★ 11

The clean slate a fresh Claude Code should start from: clean context, locked credentials, guarded egress. Isolated profile, pick-and-choose.

awesome-ai-security-engineering★ 11

Curated, vendor-neutral open-source tools, standards, benchmarks and labs for AI Security Engineering, organized by lifecycle (Discover, Design, Build, Break, Defend). From the AI Security Engineers Community.

agentic-risk-capability-framework★ 10

Website for the Agentic Risk & Capability Framework

mcp-guard★ 10

MCP middleware that blocks dangerous AI agent actions using a simple YAML config

verifiable-data-agentkit★ 10

Python SDK to verify agent tool calls, API responses, and MCP handoffs before bad outputs propagate.

open-agent-trust-registry★ 10

The open root-of-trust for agent identity on the internet. A public, federated registry of trusted attestation issuers — the agent runtimes authorized to vouch for agents acting on behalf of humans. Services verify agent attestations against this registry to determine if the issuing runtime is legitimate.

copilot-agents-guard★ 10

LLM-as-a-Judge security layer for Microsoft Copilot Studio agents

secretgate★ 10

Security proxy for AI coding agents — intercepts outbound traffic and redacts secrets before they leave your machine

AI-SPM★ 10

This opensource project dedicated to implementing Enterprise level AI-SPM. By doing so organizations can proactively protect their AI systems from threats, minimize data exposure, and maintain the trustworthiness of their AI applications (agents, mpc servers, models and more)

python-ai-guardrails-template★ 10

Engineering guardrails for AI-generated Python code. Enforces types, tests, complexity limits, security, and architecture automatically.

apoa★ 10

Agentic Power of Attorney (APOA) | Authorization infrastructure for AI agents

aura-protocol★ 10

AURA. A research frame for AI decisions, with named metrics and limits.

staplechain★ 10

Signed, In-band Annotations for Language Model Outputs

im_robot★ 10

Reverse-CAPTCHA for AI agents — verify bots, not humans. Multi-framework (React, Vue, Svelte, Web Components). Zero dependencies. TypeScript.

securecode-web★ 10

Real-World Security Conversations for AI Training

Rein-Ai★ 10

Runtime kill-switch for autonomous AI agents. Regime-aware Bayesian governor, NL policy compiler, adversarial red-team simulator. AGPL + commercial.

velar★ 10

Local Privacy Firewall for AI

fishbowl★ 10

Containerized credential auditing perimeter for AI coding agents. Wraps Codex/Claude Code in Docker, audits every credential access via eBPF.

CyberBotLLM★ 10

4 chatbots with memory made with Langchain, VertexAI and Gemini, as a cyber challenge to capture and expose RAG content.

scopeblind-gateway★ 10

Active development continues at ScopeBlind/scopeblind-gateway. - Security gateway for MCP servers. Cedar policy engine, Ed25519-signed receipts, per-tool enforcement. IETF Internet-Draft. 4 patents pending. npx protect-mcp

agbac★ 10

AGent Based Access Control (AGBAC) is is an open, vendor-neutral Internet-Draft security specification that defines how AI agents securely perform actions on behalf of humans using existing IAM technologies

safeskill★ 10

AI Skill 安全扫描:安装前查危险命令、提示注入与密钥泄露,可作 GitHub Action|Security scanner for AI agent skills

M87-Spine-lite★ 10

Stop Claude Code from doing irreversible damage. Policy-gated execution + receipts so you can ship agents without sweating production.

sentinel★ 10

MCP security scanner by Helixar

RAG-Poisoning-Lab★ 10

RAG Poisoning Lab — Educational AI Security Exercise

paperclip-plugin-avp★ 10

Trust and delegation controls for Paperclip agent teams, powered by AgentVeil Protocol.

nod★ 10

nod is a platform-agnostic, rule-based linter that ensures AI/LLM specifications contain critical security and compliance elements before any agentic or automated development begins.

A2SPA-Execution-Time-Authorization-for-Agentic-AI-Systems★ 10

A2SPA — a security protocol enforcing cryptographic authorization at execution time for agentic AI systems.

glassbox★ 10

Glass Box Framework — runtime constitutional verification for AI answers. Trust Cards with claim-level reasoning chains, formal ECS scoring, the 7-angle Glassbox Court red team, and deterministic audit logs. MCP-native.

bitgn-pac-agent-public★ 10

My solo entry for the BitGN Personal Agent Challenge 2026, first place at the Vienna hub. Defense-in-depth personal agent with layered prompt-injection hardening.

sanctuary-framework★ 10

Open source security for AI agents: an OS-enforced egress wall on macOS, and on Ubuntu 24.04 x86-64 via a separate package; keys only the operator holds; a hash-chained audit record. One command wraps Claude Code, Cursor, Hermes, OpenClaw or Cline on tested config paths. No vendor in the path. Claims trace to the Assurance Matrix.

secretctl★ 10

The simplest AI-ready secrets manager. Local-first, single-binary CLI & Desktop app with MCP integration. Never expose secrets to AI agents.

Lutren★ 10

Evidence-gated AI systems, local-first agent safety, memory continuity, MEDIOEVO tools, and GitHub Sponsors hub.

veil★ 10

Keep your secrets out of your AI coding agents. A local HTTPS proxy that swaps real credentials with placeholders and injects them at the network boundary.

systemprompt-core★ 10

AI governance infrastructure for agentic systems. Rust library behind systemprompt.io — MCP, A2A, OAuth2, audit trails, compile-time extensions. Evaluate with systemprompt-template.

deep-xpia★ 10

Benchmarking prompt injections that chain across agents

agentzt★ 10

Zero Trust for AI Agents

tene★ 10

AI-safe secret manager CLI for Claude Code, Cursor, and other AI agents. Local-first, encrypted, no cloud.

SpiralSafe★ 10

SpiralSafe is self-maintaining, inherently coherent reposystem | Documentation, code, and physical hardware in Emergent Harmony

assay★ 10

Open evidence profile for MCP tool actions. Enforce configured tool-call policies and record decisions and observations in offline-verifiable bundles. Optional Linux kernel controls. No hosted backend required.

vaara★ 10

Accountable Autonomy: open-source evidence layer that gates every AI agent tool call against your policy and writes a hash-chained record an auditor verifies offline, without trusting you. Root-agnostic; binds to TPM 2.0 / SEV-SNP when present. Your environment, no SaaS, no telemetry. AGPL-3.0.

enforra★ 10

Open source action governance SDK for AI agent tool calls.

agentaz★ 10

AgentAz™ — an open, design-time governance specification for AI agents, with 60 reference blueprints. Documents each agent's Trust Level, worst-case action, and boundaries.

postern★ 10

Credential-brokering HTTPS proxy for AI agents — brokers 1Password & Bitwarden secrets at request time, so agents call authenticated APIs without ever holding the credentials.

BoundaryAttest★ 10

Portable signed attestations for consequential agent actions crossing trust boundaries.

bastion-prompt-protection★ 10

Lightweight prompt-injection detector built for AI agents, copilots, and LLM apps. Fast, local, calibrated.

account-risk-skill★ 10

反封号.skill — Claude Code / Anthropic / OpenAI account suspension risk detection, IP cleanliness scoring, compliance preflight, and appeal-packet skill. Compliance-only; no ban evasion.

evidence-to-skill★ 10

Turn untrusted source material into compact, attributed AI skills through evidence gates and deterministic safety checks.

More in Security & AI safety