# AI Agent Security Guides, Standards, and MCP Security Canonical URL: https://pipelab.org/learn/ Description: Practical guides and open standards for AI-agent security, including MCP defense, evidence verification, compliance, and secure agent setup. Subtitle: Agent security concepts explained. ## Overview AI agents introduce security problems traditional tools were not built for. These guides explain the threats, the defenses, and where the gaps are. Each article maps to a category (IDE setup, MCP security, prompt injection, compliance, operator features, concepts) and a depth level (intro, deep dive, reference). Use the filter and search at the top of the page to narrow the list, or jump straight in: - [Start with Claude Code](/learn/claude-code-hooks/) - [Agent Firewall Guide](/agent-firewall/) ## All guides ### Tested recipes These free recipes start on Ubuntu 24.04 with curl and tar. Each uses Pipelock v3.7.1 and includes commands, observed results and evidence limits. - [Block a secret with signed community rules](/learn/block-community-secret/) - [Block synthetic patient data](/learn/block-synthetic-patient-data/) - [Block a poisoned MCP tool description](/learn/block-poisoned-mcp-tool/) - [Record signed receipts and verify them offline](/learn/record-verify-receipts/) ### IDE Setup - [Claude Code Security](/learn/claude-code-security/) - [Claude Code Hooks](/learn/claude-code-hooks/) - [Cursor AI Security](/learn/cursor-integration/) - [VS Code MCP Security](/learn/vscode-integration/) - [JetBrains MCP Security](/learn/jetbrains-integration/) - [Continue.dev MCP Security](/learn/continue/) - [Zed MCP Security](/learn/zed-integration/) ### MCP Security - [MCP Security](/learn/mcp-security/) - [MCP Vulnerabilities](/learn/mcp-vulnerabilities/) - [MCP Tool Poisoning Defense](/learn/mcp-tool-poisoning/) - [MCP Proxy](/learn/mcp-proxy/) - [MCP Gateway: Open Source Security Comparison](/learn/mcp-gateway/) - [MCP Authorization](/learn/mcp-authorization/) - [MCP Security Tools](/learn/mcp-security-tools/) - [What Is MCP?](/learn/what-is-mcp/) - [Shadow MCP](/learn/shadow-mcp/) ### Prompt Injection - [Prompt Injection Prevention](/learn/prompt-injection-network-defense/) - [LLM Prompt Injection](/learn/llm-prompt-injection/) - [Prompt Injection Detection](/learn/prompt-injection-detection/) - [Chatbot Security](/learn/chatbot-security/) ### Compliance - [AI Agent Compliance](/learn/ai-agent-compliance/) - [AI Compliance Evidence](/learn/compliance-evidence/) - [EU AI Act Compliance](/learn/eu-ai-act-compliance/) - [MITRE ATLAS Coverage](/learn/mitre-atlas/) - [NIST AI RMF Coverage](/learn/nist-ai-rmf/) - [HIPAA Security Rule Coverage](/learn/hipaa-security-rule/) - [SOC 2 Criteria Mapping](/learn/soc2-trust-services-criteria/) - [Agent Evidence Integration](/learn/agent-evidence-detection-integration/) - [Coverage Certificates](/learn/coverage-certificates/) - [OWASP MCP Top 10](/learn/owasp-mcp-top10/) - [OWASP Agentic Top 10: Coverage](/learn/owasp-agentic-top10/) - [OWASP AIVSS Scoring Context](/learn/owasp-aivss-coverage/) - [OWASP LLM Top 10: Coverage](/learn/owasp-llm-top10/) - [OWASP Agentic AI Threats: Coverage](/learn/owasp-agentic-threats/) - [SlowMist Coverage](/learn/slowmist-mcp-security-coverage/) - [Mythos-Ready Playbook](/learn/mythos-ready-playbook/) ### Evidence Standards - [Agent Evidence Level (AEL)](/learn/agent-evidence-levels/) ### Operator - [Hermes Integration](/learn/hermes-integration/) - [Codex CLI Security](/learn/codex-cli-integration/) - [Grok Build Security](/learn/grok/) - [Pi Security](/learn/pi/) - [Contain an AI Agent on Linux](/learn/pipelock-contain-guide/) - [License Setup](/learn/license-setup/) - [Pipelock Kill Switch](/learn/kill-switch/) - [Pipelock Release Notes](/learn/pipelock-release-notes/) - [Pipelock v3.7 Upgrade Guide](/learn/pipelock-v370-upgrade/) - [Launch a Command Through Pipelock](/learn/pipelock-exec/) - [Parallel Signed Receipt Chains](/learn/parallel-receipt-chains/) - [Verified Local MCP Services](/learn/verified-local-mcp-services/) - [Exact Destination Grants](/learn/destination-grants/) - [Pipelock v3.6 Upgrade Guide](/learn/pipelock-v360-upgrade/) - [Pipelock v3.3 Upgrade Guide](/learn/pipelock-v330-upgrade/) - [Pipelock v2.5 Upgrade](/learn/pipelock-v250-upgrade/) - [Pipelock v2.4 Upgrade](/learn/pipelock-v240-upgrade/) - [Progressive Enforcement](/learn/progressive-enforcement/) - [Learn-and-Lock Agent Contracts](/learn/learn-and-lock/) - [Pro Reference Deployment](/learn/pipelock-pro-reference-deployment/) - [Block Reason Headers](/learn/block-reason-headers/) - [Browser Shield](/learn/browser-shield/) - [Health Watchdog](/learn/health-watchdog/) - [Pipelock v2.3 Upgrade](/learn/pipelock-v230-upgrade/) - [AI Agent Data Redaction](/learn/ai-agent-data-redaction/) - [SSE Streaming Response Scanning](/learn/sse-streaming-response-scanning/) - [Pipelock v2.2 Upgrade](/learn/pipelock-v220-upgrade/) - [Pipelock K8s Companion Proxy](/learn/pipelock-kubernetes-companion-proxy/) - [Self-Hosted Sure + Pipelock](/learn/sure-pipelock/) - [Pipelock Session Recovery](/learn/pipelock-session-recovery/) - [Pipelock Posture Verify](/learn/pipelock-posture-verify/) - [Agent SIEM Logging](/learn/ai-agent-siem-logging/) - [Mediation Envelope Signing](/learn/mediation-envelope-signing/) - [Action Receipt Spec](/learn/action-receipt-spec/) - [AARP v0.1 Spec](/learn/aarp-spec/) - [AARP: Proves / Does Not](/learn/aarp-what-it-proves/) - [AARP Claims Dictionary](/learn/aarp-claims-dictionary/) - [Flight Recorder: AI Agent Audit Log](/learn/flight-recorder/) - [Pipelock Detection Rules](/learn/community-rules/) - [Canary Tokens: Synthetic Secret Detection](/learn/canary-tokens/) ### Concepts - [Agent Security Tool Profiles](/learn/agent-security-tool-profiles/) - [AI Agent Security Categories](/learn/ai-agent-security-categories/) - [Agent Firewall vs Guardrails vs Sandbox](/learn/agent-security-model/) - [What Pipelock Claims and Doesn't](/learn/non-bypass-doctrine/) - [Known Limitations](/learn/known-limitations/) - [AI Agent Security](/learn/ai-agent-security/) - [Agent Security Tools](/learn/ai-agent-security-tools/) - [AI Egress Proxy](/learn/ai-egress-proxy/) - [Denial of Wallet](/learn/denial-of-wallet/) - [Pipelock Performance](/learn/performance/) - [Open Source AI Firewall](/learn/open-source-ai-firewall/) - [Secure AI Agents](/learn/agent-egress-security/) - [Cloudflare Sandboxes + Pipelock](/learn/cloudflare-sandboxes-pipelock/) - [Generative AI Firewall](/learn/generative-ai-firewall/) - [AI Runtime Security](/learn/ai-runtime-security/) - [LLM Security](/learn/llm-security/) - [Secure Agent Deployment](/learn/secure-ai-agent-deployment/) - [AI Agent Data Loss Prevention](/learn/ai-agent-data-loss-prevention/) ## Pages - [Block a poisoned MCP tool description](https://pipelab.org/learn/block-poisoned-mcp-tool/): Wrap a small MCP test server with Pipelock on Ubuntu, compare a clean tool list with a poisoned description, and verify the blocked response. - [Block a fake secret with signed community rules](https://pipelab.org/learn/block-community-secret/): Install signed Pipelock community rules on Ubuntu, block a planted fake secret in an outbound HTTP request, and read the verdict explanation. - [Block synthetic patient data with Pipelock](https://pipelab.org/learn/block-synthetic-patient-data/): Install the signed healthcare rule bundle on Ubuntu, block a synthetic medical record number in an HTTP request, and inspect the verdict. - [Record signed receipts and verify them offline](https://pipelab.org/learn/record-verify-receipts/): Record a blocked request with Pipelock on Ubuntu, verify its signed receipt with the standalone verifier, and reject a tampered evidence copy. - [Codex CLI Security: Wrap MCP Servers and Route Egress](https://pipelab.org/learn/codex-cli-integration/): Codex CLI security guide. Wrap Codex MCP servers with pipelock codex install, route proxy-aware commands through Pipelock, and see what each layer covers. - [Pipelock Release Notes: Release Posts and Upgrade Guides](https://pipelab.org/learn/pipelock-release-notes/): Pipelock release posts, newest first, with dates, what changed in each release, and the matching upgrade guides. The GitHub changelog covers every version. - [Exact Destination Grants in Pipelock](https://pipelab.org/learn/destination-grants/): Authorize one TCP host and port with Pipelock destination grants while retaining content scanning, metadata protection, and fail-closed DNS checks. - [Parallel Signed Receipt Chains in Pipelock](https://pipelab.org/learn/parallel-receipt-chains/): Configure parallel signed receipt chains in Pipelock, verify the complete group, handle restart and recovery seals, and retain honest end proofs. - [Launch a Command Through Pipelock](https://pipelab.org/learn/pipelock-exec/): Use pipelock exec to check a running proxy and launch cooperative commands with proxy and CA settings. Use containment for enforced isolation. - [Pipelock v3.7 Upgrade Guide](https://pipelab.org/learn/pipelock-v370-upgrade/): Upgrade Pipelock v3.7 with complete-chain verification, filesystem containment grants, local MCP identities, and stricter request inspection. - [Verify Local MCP Service Identity](https://pipelab.org/learn/verified-local-mcp-services/): Verify local MCP services by kernel-observed owner and pinned program inputs, then bind reviewed credential-request acknowledgments to that identity. - [Pipelock v3.6 Upgrade Guide: Breaking Changes and Checks](https://pipelab.org/learn/pipelock-v360-upgrade/): Pipelock v3.6 upgrade guide: run the binary check, fix refused host patterns and expiries, update receipt and Scan API readers, and reinstall containment. - [Contain an AI Agent on Linux with Pipelock](https://pipelab.org/learn/pipelock-contain-guide/): Contain an AI agent on one Linux host with pipelock contain. Install, grant a workspace, launch, watch the screen, and verify the signed change statement. - [Pi Security: Route the Coding Agent Through Pipelock](https://pipelab.org/learn/pi/): Route the Pi coding agent through Pipelock with pipelock pi install. Set a named proxy listener, preview with --dry-run, verify, and undo with remove. - [Grok Build Security: Forward Proxy and Manual MCP Wrap](https://pipelab.org/learn/grok/): Grok Build security guide. Point the CLI through Pipelock with HTTPS_PROXY, wrap MCP servers by hand, and see what TLS interception adds. - [HIPAA Security Rule Coverage: Pipelock Mapping](https://pipelab.org/learn/hipaa-security-rule/): HIPAA Security Rule coverage from Pipelock: selected technical safeguards, runtime evidence, PHI disclosure limits, and the controls outside the proxy. - [MITRE ATLAS Coverage: Pipelock Technique Mapping](https://pipelab.org/learn/mitre-atlas/): MITRE ATLAS coverage from Pipelock: fourteen mapped techniques, the runtime controls behind each mapping, and the limits the proxy cannot own. - [NIST AI RMF Coverage: Pipelock Control Mapping](https://pipelab.org/learn/nist-ai-rmf/): NIST AI RMF coverage from Pipelock: GOVERN, MAP, MEASURE, MANAGE, and Generative AI mappings with the runtime limits stated plainly. - [SOC 2 Trust Services Criteria: Pipelock Mapping](https://pipelab.org/learn/soc2-trust-services-criteria/): SOC 2 Trust Services Criteria coverage from Pipelock: security, availability, processing integrity, confidentiality, privacy, and stated runtime limits. - [The Pipelock Kill Switch: Six Sources, One Deny-All](https://pipelab.org/learn/kill-switch/): How the Pipelock kill switch works: six OR-composed sources, port isolation so agents cannot self-deactivate, and automation over the API. - [Denial of Wallet: AI Agent Cost Control and LLM Cost Abuse](https://pipelab.org/learn/denial-of-wallet/): Denial of wallet is LLM cost abuse from looping agents. Learn why session caps fail and how subject-keyed AI agent budgets work. - [Pipelock v3.3 Upgrade Guide: Breaking Changes and Checks](https://pipelab.org/learn/pipelock-v330-upgrade/): Pipelock v3.3 upgrade guide: run the new binary check, remove failed budget fields, review subject budgets, receipts, rules, and SNI. - [AI Agent Security Event Logging: Ship Agent Events to Your SIEM](https://pipelab.org/learn/ai-agent-siem-logging/): Ship AI agent security events to your SIEM: blocked requests, prompt injection hits, and kill-switch activations over webhook, syslog, or OTLP. - [Coverage Certificates: Signed Proof of What Your Agent Firewall Mediated](https://pipelab.org/learn/coverage-certificates/): Coverage certificates: signed, offline-verifiable attestations of the AI agent sessions Pipelock mediated over a bounded window, exclusions visible. - [The Pipelock Operator Console](https://pipelab.org/learn/operator-console/): The Pipelock operator console: a read-only view over signed agent evidence, exemptions, budgets, and fleet state, with offline-verifiable proof. - [Why Pipelock Is an Egress Agent Firewall, Not an Inbound WAF](https://pipelab.org/learn/egress-vs-inbound-agent-firewall/): Why Pipelock is an egress agent firewall, not an inbound WAF, and why blocking the outbound effect beats filtering prompts at the door for AI agents. - [Agent Evidence Level (AEL): Grading AI-Agent Audit Evidence](https://pipelab.org/learn/agent-evidence-levels/): Agent Evidence Level is an open standard for grading AI-agent audit evidence by what an independent party can verify offline. - [Agent Firewall, Guardrails, Sandbox: Defining the Category](https://pipelab.org/learn/agent-security-model/): Agent firewall, guardrails, and sandbox compared as four security functions: prevention, detection, containment, and evidence. What each does and misses. - [Claude Code Security: Harden the Agent Runtime](https://pipelab.org/learn/claude-code-security/): Claude Code security beyond built-in review. Add egress inspection, outbound DLP, MCP response scanning, and containment you can verify. - [Pipelock Known Limitations: What It Does Not Catch](https://pipelab.org/learn/known-limitations/): A standing register of what Pipelock does not catch: process-level bypass, non-HTTP exfiltration, covert channels, and semantic prompt injection. - [Progressive Enforcement for AI Agents](https://pipelab.org/learn/progressive-enforcement/): Progressive enforcement for AI agents: observe, baseline, shadow, then enforce. Roll out a blocking agent firewall with signed, reversible receipts. - [What Pipelock Claims, and What It Doesn't](https://pipelab.org/learn/non-bypass-doctrine/): Pipelock scans routed traffic, supports Linux containment with live verification, and signs mediated decisions. These do not prove that no bypass occurred. - [AARP Claims Dictionary](https://pipelab.org/learn/aarp-claims-dictionary/): AARP claims dictionary: verified claims, reserved vocabulary, does_not_assert limits, and overclaim-risk warnings for receipt appraisal. - [AARP v0.1: The Agent Action Receipt Profile (Spec)](https://pipelab.org/learn/aarp-spec/): AARP v0.1 spec: a signed assurance profile over Pipelock receipts. Claim-set by axis, JCS canonicalization, parallel signatures, X.509-SVID identity. - [What an AARP Receipt Proves, and What It Does Not](https://pipelab.org/learn/aarp-what-it-proves/): What a verified AARP receipt proves and does not prove. Binary-enforced versus deployment-claimed, and the trust assumptions a relying party must pin. - [Hermes Agent Integration: Hook-Based Scanning for Non-MCP Agents](https://pipelab.org/learn/hermes-integration/): Bridge Hermes Agent into Pipelock's scanner pipeline with a hook plugin. Install, verify, and roll back the integration with one command each. - [AI Redaction Placeholders: What They Mean](https://pipelab.org/learn/redaction-placeholders/): Decode AI redaction placeholders like , , and : what each token means, what tool emits it, and why. - [Request Policy: Allow and Deny Individual API Operations](https://pipelab.org/learn/request-policy/): Pipelock request policy allows or denies individual outbound API operations: GraphQL mutations, JSON batch sub-requests, and admin calls, not just hosts. - [Pipelock Pro Reference Deployment](https://pipelab.org/learn/pipelock-pro-reference-deployment/): Pipelock Pro reference deployment: per-agent budgets, source-CIDR routing, cross-agent federation, signed receipts. The architecture, not the marketing. - [Agent Security Tool Profiles: Procurement Evidence](https://pipelab.org/learn/agent-security-tool-profiles/): Receipt-scoring profiles for agent security tools: signed, offline-verifiable evidence buyers can reproduce against the public corpus. - [Agent Action Receipts: Signed Evidence for What an AI Agent Did](https://pipelab.org/learn/agent-action-receipts/): Agent action receipts: Ed25519-signed records of AI agent network and tool actions, chained by hash, verifiable offline by any third party. - [Agent Security Control Layers: Sandboxing, Identity, MCP Gateway, Egress Inspection](https://pipelab.org/learn/agent-security-control-layers/): Four-layer map of AI agent security: process sandboxing, identity governance, MCP gateway mediation, and egress inspection with signed receipts. - [Verifiable Egress Control: Binary-Enforced Mediation with Signed Evidence](https://pipelab.org/learn/verifiable-egress-control/): Verifiable Egress Control combines network-enforced AI agent egress with signed action receipts auditors and procurement teams can verify offline. - [Verify a Pipelock Receipt: Copy-Paste Demo](https://pipelab.org/learn/verify-a-receipt/): Verify Pipelock Ed25519 action receipts with Go, TypeScript, Rust, or Python verifier CLIs against the public conformance corpus. - [Pipelock v2.5 Upgrade Guide](https://pipelab.org/learn/pipelock-v250-upgrade/): Pipelock v2.5 upgrade guide: verifier rollout, containment lifecycle, strict federation, MCP integrity, and installer checks. - [Audit Packet Threat Model: What Verified Receipts Prove](https://pipelab.org/learn/audit-packet-threat-model/): Threat model for the Pipelock Audit Packet. Receipt provenance, self-consistent verdicts, signer-key pinning, what verified evidence does not prove. - [Browser Shield: Defensive Response Rewriting](https://pipelab.org/learn/browser-shield/): Browser Shield sanitizes supported HTML and SVG markup before agent delivery. Existing inline scripts and standalone JavaScript stay unchanged. - [Continue.dev MCP Security: Scanning with Pipelock](https://pipelab.org/learn/continue/): Continue.dev MCP security guide. Wrap each MCP server in Continue's config through pipelock mcp proxy for DLP, injection, and tool-poisoning scanning. - [Pipelock Performance](https://pipelab.org/learn/performance/): Latency, cold-start, and memory overhead numbers for Pipelock across HTTP, SSE, MCP stdio, WebSocket, and tool-call chains. Reproducible. - [Zed MCP Security: Scanning with Pipelock](https://pipelab.org/learn/zed-integration/): Zed MCP security guide. Install Pipelock to scan every context_server tool call in Zed stable, Zed Preview, and Flatpak Zed. - [Pipelock License Setup](https://pipelab.org/learn/license-setup/): Pipelock license setup guide. Install your Pro or Founding Pro token, verify it loaded, and start using premium features. - [MCP Runtime Security: Live-Traffic Defenses for AI Agents](https://pipelab.org/learn/mcp-runtime-security/): MCP runtime security covers live-traffic defenses pre-deploy scanners miss: tool poisoning in responses, rug-pull drift, chain attacks, fail-closed. - [AI Agent Regulatory Controls and Evidence Hub](https://pipelab.org/learn/ai-agent-regulatory-controls/): AI agent regulatory controls: shipped Pipelock Assess mappings for EU AI Act, NIST AI RMF, SOC 2 and OWASP, plus DORA, NIS2, ISO 42001 and Colorado notes. - [AI Agent Action Receipts: Verify What an Agent Did Offline](https://pipelab.org/learn/what-did-my-agent-do/): Live demo for AI coding agents: Pipelock blocks a malicious MCP response and emits an Ed25519-signed action receipt verified offline. - [Agent Egress Control: GitHub Action Setup](https://pipelab.org/learn/agent-egress-control/): Pipelock Agent Egress Control GitHub Action setup guide. Kernel-enforced containment, signed Audit Packets, pinning options, offline receipt verification. - [Compliance Evidence Substrate for AI Agents](https://pipelab.org/learn/compliance-evidence-substrate/): Pipelock as evidence substrate: EU AI Act Article 12 and NIST AI RMF mapped to signed action receipts, scanner verdicts, and audit packets. - [Skill Supply Chain Security: Poisoning Attacks Against npx skills, vercel-labs/skills, and the Open Skills Ecosystem](https://pipelab.org/learn/skill-supply-chain-security/): Skill supply chain security for SKILL.md poisoning: where Pipelock scans, what v2.5 catches, and what egress controls stop. - [Cloudflare AI Gateway: Pricing, Features, Limits & Alternatives (2026)](https://pipelab.org/learn/cloudflare-ai-gateway/): Cloudflare AI Gateway pricing: caching, rate limits, analytics, and DLP are free. Guardrails are billed; log limits and charges depend on the logging plan. - [Block Reason Headers](https://pipelab.org/learn/block-reason-headers/): Pipelock v2.4 X-Pipelock-Block-Reason header reference: vocabulary, severity, retry hints, transports, and agent integration patterns. - [Pipelock Health Watchdog](https://pipelab.org/learn/health-watchdog/): Pipelock v2.4 health watchdog reference: the /health endpoint, subsystem map, hybrid passive plus active probe, and Kubernetes liveness pattern. - [Learn-and-Lock Agent Contracts](https://pipelab.org/learn/learn-and-lock/): Learn-and-lock turns observed AI agent traffic into signed Pipelock contracts, then tests them in shadow mode before enforcement. - [Pipelock v2.4 Upgrade Guide](https://pipelab.org/learn/pipelock-v240-upgrade/): Pipelock v2.4 upgrade guide: learn-and-lock rollout, block reason headers, inbound envelope verification, Gemini redaction, and health checks. - [AI Agent Security Categories: What Each One Catches and Misses](https://pipelab.org/learn/ai-agent-security-categories/): A buyer's map of the six AI agent security categories. What each one controls, what each one misses, and how they stack. - [Securing Claude Code Against Secret Exfiltration](https://pipelab.org/learn/securing-claude-code-against-secret-exfiltration/): How Claude Code can leak credentials through tool calls, MCP servers, and shell, plus the runtime defenses that catch each path before traffic leaves. - [Preventing SSRF in AI Agents: Attack Vectors and Defenses](https://pipelab.org/learn/preventing-ssrf-in-ai-agents/): SSRF against AI agents: cloud metadata, private CIDRs, DNS rebinding, encoded IPs, parser-differential gaps, and defenses that work. - [Self-Hosted Sure with Pipelock: Secure AI Assistant Egress](https://pipelab.org/learn/sure-pipelock/): Self-hosted Sure guide for enabling Pipelock in Helm, gating external AI assistant egress, and validating the chart guard. - [Agent Evidence: SIEM and Detection Integration](https://pipelab.org/learn/agent-evidence-detection-integration/): Pipelock action receipts integrate with SIEM, audit, and LLM detection pipelines. Tamper-evident JSONL, hash-chained, Ed25519-signed. - [AI Agent Data Redaction](https://pipelab.org/learn/ai-agent-data-redaction/): Pipelock redacts AI agent secrets in flight using class-preserving placeholders across Anthropic, OpenAI, Gemini, and custom JSON providers. - [Pipelock v2.3 Upgrade Guide](https://pipelab.org/learn/pipelock-v230-upgrade/): Pipelock v2.3 upgrade guide: drop-in upgrade from v2.2.x, plus how to enable class-preserving redaction and generic SSE streaming scanning. - [SSE Streaming Response Scanning](https://pipelab.org/learn/sse-streaming-response-scanning/): Pipelock v2.3.0 streams every SSE response with per-event DLP and prompt injection scanning. OpenAI, Anthropic, Kilo Gateway, any LLM SSE. - [LLM Security: A Practitioner's Guide to Protecting Large Language Models and AI Agents](https://pipelab.org/learn/llm-security/): LLM security covers prompt injection, data leaks, tool poisoning, and agent runtime attacks. Practitioner guide to threats and where defenses fit. - [AI Agent Data Loss Prevention: How DLP Works for Agents (and Where It Breaks)](https://pipelab.org/learn/ai-agent-data-loss-prevention/): AI agent DLP guide: where agents leak credentials and PII, what catches it at the network layer, and the open-source self-hosted approach. - [Chatbot Security: Risks, Defenses, and Where Network Controls Fit](https://pipelab.org/learn/chatbot-security/): Chatbot security guide: credential leaks, prompt injection, jailbreaks, oversharing, and the network-layer controls that catch what the model misses. - [What Is MCP? Model Context Protocol Explained for AI Agents](https://pipelab.org/learn/what-is-mcp/): Plain-language guide to Model Context Protocol (MCP): what it is, who built it, how it works, and how it differs from function calling and RAG. - [Mediation Envelope Signing and Inbound Verification](https://pipelab.org/learn/mediation-envelope-signing/): Pipelock mediation envelope guide: RFC 9421 signing, inbound verification, SPIFFE actors, and the well-known directory for cross-org federation. - [Pipelock Kubernetes Companion Proxy](https://pipelab.org/learn/pipelock-kubernetes-companion-proxy/): Pipelock Kubernetes companion proxy guide: generate an enforced proxy Deployment, Service, NetworkPolicies, and bound identity from a workload manifest. - [Pipelock Posture Verify Guide](https://pipelab.org/learn/pipelock-posture-verify/): Pipelock posture verify guide: validate signed posture capsules, enforce policy thresholds, and use exit codes separating integrity from policy failures. - [Pipelock Session Recovery](https://pipelab.org/learn/pipelock-session-recovery/): Pipelock session recovery guide: use inspect, explain, release, terminate, and recover to handle airlocked sessions without guessing at proxy state. - [Pipelock v2.2 Upgrade Guide](https://pipelab.org/learn/pipelock-v220-upgrade/): Pipelock v2.2 upgrade guide: strict YAML validation, rollout checks, and the operator steps to verify config before you cut traffic over. - [AI Runtime Security: Defending AI Agents and Models at Execution Time](https://pipelab.org/learn/ai-runtime-security/): AI runtime security covers model, agent, and infrastructure threats at execution time: prompt injection, tool misuse, egress exfiltration, and defenses. - [Generative AI Firewall: What It Is and When You Need One](https://pipelab.org/learn/generative-ai-firewall/): Generative AI firewall guide to prompt filtering, output scanning, agent egress control, vendors, and where open-source tools fit. - [Cloudflare Sandboxes and Pipelock: Two-Layer Egress Control for AI Agents](https://pipelab.org/learn/cloudflare-sandboxes-pipelock/): Cloudflare Sandboxes provides agent isolation and domain filtering. Pipelock adds content scanning for credentials, injection, and tool poisoning. - [The Mythos-Ready Playbook: Runtime Controls for the AI Vulnerability Storm](https://pipelab.org/learn/mythos-ready-playbook/): CSA/SANS/OWASP Mythos-Ready playbook mapped to runtime controls: egress filtering, agent containment, and machine-speed response. - [MCP Authorization: Access Control for AI Agent Tool Calls](https://pipelab.org/learn/mcp-authorization/): MCP authorization controls which agents access which tools. Covers OAuth 2.1, scopes, tool-level RBAC, confused deputy, and audit patterns. - [OWASP MCP Top 10 (MCP01:2025 through MCP10:2025): Risks and Practical Defenses](https://pipelab.org/learn/owasp-mcp-top10/): OWASP MCP Top 10 (MCP01:2025-MCP10:2025): each risk category explained with the scanner, gateway, inspection, and audit controls that stop it. - [Shadow MCP: Find and Lock Down Rogue MCP Servers](https://pipelab.org/learn/shadow-mcp/): Shadow MCP is the unauthorized MCP connectivity hiding in your codebase. How to find rogue MCP servers, score the risk, and enforce policy at runtime. - [AI Agent Security Tools: Scanners, Firewalls, and Gateways](https://pipelab.org/learn/ai-agent-security-tools/): AI agent security tools range from static scanners to runtime firewalls. Compare what each layer catches and pick the right stack for your agents. - [AI Agent Compliance: Audit Logs, Policy, and Evidence](https://pipelab.org/learn/ai-agent-compliance/): AI agent compliance needs audit logs, runtime policy controls, and signed evidence. Map agent behavior to SOC 2, EU AI Act, and OWASP frameworks. - [AI Egress Proxy: Route and Inspect All Agent Traffic](https://pipelab.org/learn/ai-egress-proxy/): How an AI egress proxy works: route agent HTTP, WebSocket, and MCP traffic through one inspection point, deployed via env var or companion proxy. - [MCP Gateway Open Source vs Commercial: Comparison and How to Choose](https://pipelab.org/learn/mcp-gateway/): MCP gateway guide comparing open source and commercial options, routing, auth, content inspection, proxy differences, and security controls. - [Best MCP Security Tools 2026: Scanners, Gateways, Proxies Compared](https://pipelab.org/learn/mcp-security-tools/): Best MCP security tools for 2026 compared: pre-deploy scanners, runtime proxies, and MCP gateways, with what each one catches and misses. - [AI Firewall vs LLM Firewall vs Agent Firewall: Open-Source Options (2026)](https://pipelab.org/learn/open-source-ai-firewall/): Free, open source, self-hosted AI firewall options compared: AI firewall vs LLM firewall vs agent firewall, with Pipelock, LlamaFirewall, and NeMo. - [Secure AI Agent Deployment: Pre-Launch to Production](https://pipelab.org/learn/secure-ai-agent-deployment/): Secure AI agent deployment starts before launch. Isolate credentials, restrict network access, inspect tool traffic, and log every decision. - [Pipelock Action Receipt Format (Implementation Spec)](https://pipelab.org/learn/action-receipt-spec/): Pipelock signed receipt formats: Ed25519 ActionReceipt v1 for proxy decisions, plus EvidenceReceipt v2 for v2.4 contract lifecycle and shadow evidence. - [AI Agent Security & Hardening: 2026 Checklist Mapped to OWASP](https://pipelab.org/learn/ai-agent-security/): AI agent security and hardening checklist mapped to OWASP: identity, scoped tokens, sandbox, egress allowlist, approvals, logging, and kill switch. - [LLM Prompt Injection: What It Is and Why It Matters for AI Agents](https://pipelab.org/learn/llm-prompt-injection/): LLM prompt injection explained. How attackers hijack AI agents through malicious text in tool responses, web pages, and MCP servers. Defenses included. - [MCP Proxy: How It Scans MCP Traffic Bidirectionally](https://pipelab.org/learn/mcp-proxy/): An MCP proxy sits between agents and servers, inspecting tool descriptions, arguments, and responses for injection, credentials, and rug-pulls. - [MCP Tool Poisoning: Detection and Runtime Defense](https://pipelab.org/learn/mcp-tool-poisoning/): MCP tool poisoning hides malicious instructions in tool metadata. Catch rug-pulls and block unsafe tool changes at runtime with proxy-layer defense. - [MCP Vulnerabilities: Known Risks and Defenses](https://pipelab.org/learn/mcp-vulnerabilities/): MCP vulnerabilities mapped: tool poisoning, rug-pulls, credential theft, SSRF, prompt injection, session hijacking. Runtime defenses for each risk. - [Prompt Injection Detection: Techniques and Tools](https://pipelab.org/learn/prompt-injection-detection/): Prompt injection detection techniques for AI agents. Pattern matching, ML classifiers, normalization pipelines, and how to combine detection layers. - [AI Compliance Evidence: Signed Reports](https://pipelab.org/learn/compliance-evidence/): AI compliance evidence from Pipelock. Maps runtime controls to seven frameworks and generates signed bundles. SARIF integrates with GitHub Code Scanning. - [Canary Tokens: Synthetic Secret Detection](https://pipelab.org/learn/canary-tokens/): Canary tokens for AI agent security. Plant synthetic secrets in Pipelock that trigger alerts when an agent attempts to exfiltrate them. - [Flight Recorder: AI Agent Audit Log](https://pipelab.org/learn/flight-recorder/): AI agent audit log with hash-chained, tamper-evident entries. Pipelock's flight recorder writes DLP-redacted JSONL with Ed25519 signed checkpoints. - [JetBrains MCP Security: Scanning with Pipelock](https://pipelab.org/learn/jetbrains-integration/): JetBrains MCP security guide. Install Pipelock to scan MCP connections in IntelliJ IDEA, PyCharm, WebStorm, GoLand, and Junie agents. - [OWASP AIVSS: Scoring Context and Pipelock Evidence](https://pipelab.org/learn/owasp-aivss-coverage/): OWASP AIVSS scoring context: risk amplification and mitigation factors. Pipelock Assess v3.7.1 does not ship an AIVSS mapping or score reduction. - [Pipelock Detection Rules: Community Collection](https://pipelab.org/learn/community-rules/): Pipelock detection rules from the community. 28 signed rules covering DLP patterns, MCP tool poisoning, and prompt injection scanning. - [SlowMist MCP Security: Pipelock Coverage](https://pipelab.org/learn/slowmist-mcp-security-coverage/): Pipelock covers 10 of 19 SlowMist MCP security test cases fully, 8 partially, with 1 out of scope. Full item-by-item breakdown. - [VS Code MCP Security: Scanning with Pipelock](https://pipelab.org/learn/vscode-integration/): VS Code MCP security guide. Install Pipelock to scan MCP server traffic for credential leaks, prompt injection, and tool poisoning before execution. - [Pipelock Hooks for Claude Code: Setup Guide](https://pipelab.org/learn/claude-code-hooks/): Install Pipelock hooks for Claude Code in one command. Scans Bash, WebFetch, Write, Edit, and MCP tool calls for credential leaks and injection. - [OWASP Agentic AI Threats: Pipelock Coverage](https://pipelab.org/learn/owasp-agentic-threats/): OWASP Agentic AI Threats coverage from Pipelock's published v3.7.1 documentation mapping: T1-T15 ratings, recognized gaps, and Assess limits. - [OWASP Agentic Top 10: Pipelock Coverage](https://pipelab.org/learn/owasp-agentic-top10/): OWASP Top 10 for Agentic Applications 2026 (ASI01-ASI10) in Pipelock's shipped mapping: four risks covered, five partial, one not covered, with limits. - [OWASP LLM Top 10: Pipelock Coverage](https://pipelab.org/learn/owasp-llm-top10/): OWASP LLM Top 10 coverage from Pipelock's published v3.7.1 documentation mapping, with ratings, out-of-scope risks, and Assess report limits. - [EU AI Act Compliance: Article 26 Deployer Obligations and 6-Month Log Retention](https://pipelab.org/learn/eu-ai-act-compliance/): EU AI Act Article 26(6): deployers of high-risk AI must keep automatically generated logs for at least six months. Who keeps them and how agents comply. - [Cursor AI Security: Scanning Hooks with Pipelock](https://pipelab.org/learn/cursor-integration/): Cursor AI security guide. Install Pipelock hooks to block credential exfiltration, reverse shells, and dangerous commands before they execute. - [Agent Egress Security: How AI Agents Leak Credentials and How to Stop It](https://pipelab.org/learn/agent-egress-security/): How AI agents leak API keys and tokens through HTTP, DNS, and MCP tool calls, why classic DLP misses it, and the egress controls that stop it. - [MCP Security: Best Practices Checklist (2026)](https://pipelab.org/learn/mcp-security/): MCP security best practices checklist for 2026, mapped to the OWASP MCP Top 10: tool poisoning, secret leaks, SSRF, runtime defenses, and audit. - [Prompt Injection Prevention at the Network Layer](https://pipelab.org/learn/prompt-injection-network-defense/): Prompt injection prevention beyond the model layer. Network scanning catches injection in HTTP and MCP traffic before it reaches the AI agent.