# GuardionAI > AI Security Gateway for production AI systems ## About GuardionAI provides an AI Security Gateway that intercepts, inspects, and enforces policies on LLM and MCP traffic in real time. It protects AI agents from prompt injection, data exfiltration, and supply chain attacks with sub-millisecond guardrails. Founded by Rafael Sandroni, ex-Apple Engineer and Entrepreneur First alumnus. ## Social - Company LinkedIn: https://www.linkedin.com/company/guardion-ai - Founder LinkedIn: https://linkedin.com/in/rafaelsandroni ## Product - [Home](https://guardion.ai) - [Pricing](https://guardion.ai/pricing) - [Request a Demo](https://guardion.ai/request-demo) - [Contact](https://guardion.ai/contact) - [Playground](https://guardion.ai/playground) - [FAQ](https://guardion.ai/faq) - [Product Platform](https://guardion.ai/product): AI Security Gateway platform overview - [Security Gateway](https://guardion.ai/product/security-gateway): Inline control point for every agent and MCP call. - [Data Loss Prevention for Agents](https://guardion.ai/product/data-loss-prevention): Anonymize PII and protect secrets before they leave your org. - [Runtime Guardrails for Agents](https://guardion.ai/product/runtime-guardrails): Detect prompt injection, jailbreaks, and unsafe actions. - [Claude Code Plugins](https://guardion.ai/product/claude-code-plugins): Govern coding agents on the developer machine, EDR-style. - [Detection & Incident Response](https://guardion.ai/product/detection-incident-response): A centralized hub to investigate and respond to agent incidents. ## Solutions - [Solutions](https://guardion.ai/solutions): Guardion by use case, team, and industry - [Stop data exfiltration](https://guardion.ai/solutions/stop-data-exfiltration): Block data exfiltration from autonomous agents and MCP tools at runtime, before the action executes. - [Govern coding agents](https://guardion.ai/solutions/govern-coding-agents): Govern Claude Code, Copilot, and Cursor on the developer machine — commands, secrets, and risky dependencies — in real time. - [Chatbots](https://guardion.ai/solutions/chatbots): Govern customer-facing chatbots — block prompt injection, jailbreaks, and off-policy output, and redact PII before it reaches a customer. - [Support chatbots](https://guardion.ai/solutions/support-chatbots): Keep support and CX chatbots on-policy and compliant — prevent unauthorized advice, off-topic drift, and PII leakage, with a full audit trail. - [Agentic workflows](https://guardion.ai/solutions/agentic-workflows): Govern multi-step agentic workflows — observe every step, enforce policy on each tool call, and catch intent drift before it compounds. - [AI copilots & assistants](https://guardion.ai/solutions/ai-copilots): Govern internal AI copilots and assistants — enforce least-privilege actions, redact secrets and PII, and audit every access. - [Security & GRC](https://guardion.ai/solutions/security-grc): Give security and GRC teams control and audit over agent actions — without becoming the blocker. - [Platform & AI Engineering](https://guardion.ai/solutions/platform-ai-engineering): Zero-instrumentation runtime governance for platform and AI engineering teams shipping agents to production. - [Fintech & Banking](https://guardion.ai/solutions/fintech): Govern AI agents in fintech and banking with multilingual guardrails, PII anonymization, and LGPD/GDPR-ready audit trails. - [Healthcare](https://guardion.ai/solutions/healthcare): Protect PHI and govern AI agent actions for healthcare with inline anonymization, a security vault, and HIPAA-grade audit trails. ## Guides & Concepts - [Docs](https://guardion.ai/docs): Product documentation - [Guides & Concepts](https://guardion.ai/docs/guides): How Guardion's guard models, policy engine, and observability work - [Guard models](https://guardion.ai/docs/guides/guard-models): Guardion runs a set of purpose-built detection models over every message, tool definition and response — covering prompt attacks, content safety and sensitive data — and returns a calibrated score and labels you can act on. - [Prompt defense](https://guardion.ai/docs/guides/prompt-defense): Detects adversarial inputs that try to subvert your system prompt or policies — prompt injection and jailbreaks — plus optional bot and spam abuse signals. - [Moderation model](https://guardion.ai/docs/guides/moderation): Classifies content against standard safety categories — hate, harassment, violence, sexual content, self-harm and illicit/dangerous activity — for both user inputs and model outputs. - [PII & DLP model](https://guardion.ai/docs/guides/pii-dlp): Detects personal data and secrets in text and tool arguments, and can redact them in place — reversibly — so sensitive values never reach the model, your logs, or a downstream tool. - [Policy engine](https://guardion.ai/docs/guides/policy-engine): How Guardion turns raw detector results and accumulated session context into a single, explainable decision — allow, flag, deny or modify. - [Decision matrix & examples](https://guardion.ai/docs/guides/decision-matrix): The full mapping from guardrail actions and policy action to a decision, with worked examples for every combination. - [Session risk & rules](https://guardion.ai/docs/guides/session-risk): How Guardion accumulates risk across a conversation and how declarative rules use that context to escalate a decision beyond a single turn. - [Telemetry & observability](https://guardion.ai/docs/guides/observability): How Guardion records every evaluation as structured telemetry — sessions, traces and spans — so you can audit decisions, investigate incidents and export the data. ## API Reference Base URLs: https://api.guardion.ai/v1 (Guard) and https://gateway.guardion.ai (Gateway). - [Introduction](https://guardion.ai/docs/api-reference/introduction): The Guardion API runs AI-security guardrails over any LLM or agent interaction — detecting prompt injection, jailbreaks, PII, toxic content and tool-poisoning — and lets you export the resulting telemetry. - [Authentication](https://guardion.ai/docs/api-reference/authentication): Guardion authenticates every request with an API key. How you pass it differs between Guard and the Gateway. - [Response codes](https://guardion.ai/docs/api-reference/response-codes): Guard uses conventional HTTP status codes. The Gateway adds two intentional guardrail status codes (246 and 446) you can branch on. - [Rate limits](https://guardion.ai/docs/api-reference/rate-limits): Rate limits are applied per API key. - [FAQ](https://guardion.ai/docs/api-reference/faq): Common questions about evaluating, enforcing, configuring and exporting with Guardion. - [Enforce policies](https://guardion.ai/docs/api-reference/guard): Run the detector pipeline for one or more policies against a list of messages. Returns whether the content is flagged, whether it should be denied, any redactions/corrections, and a per-detector breakdown. - [CLI agent hooks](https://guardion.ai/docs/api-reference/hooks-events): Receives hook-event payloads from coding-agent runtimes (Claude Code and similar). Each event is normalized into messages and run through the same detector pipeline as /v1/guard. Monitors by default and can block at true pre-action gates (PreToolUse, UserPromptSubmit). Fail-open: never crashes the agent. - [Export logs](https://guardion.ai/docs/api-reference/logs): Returns paginated interaction logs (one row per evaluated request) with the messages and per-detector eval results. - [Export findings](https://guardion.ai/docs/api-reference/findings): Returns paginated, de-duplicated security findings — aggregated threats with their occurrences, guardrail evidence, and affected sessions. Answers what happened, where, and how. Note: findings live under the /v2 prefix. - [Redact PII](https://guardion.ai/docs/api-reference/pii-redact): Detects PII, replaces it with stable tokens, and stores the originals in a time-limited vault. Returns a vault_id you can later pass to /unredact to restore the originals. - [Restore PII](https://guardion.ai/docs/api-reference/pii-unredact): Restores original PII from tokens using the vault_id returned by /redact. - [Chat completions](https://guardion.ai/docs/api-reference/chat-completions): OpenAI-compatible chat completions through the Guardion Gateway. Same request and response shape as the OpenAI Chat Completions API; guardrails run inline. If a guardrail blocks under a block config, the gateway returns the Guardion verdict instead of the model output. ## Runtime Guardrail & LLM Security Benchmark - [Leaderboard](https://guardion.ai/leaderboard): Runtime guardrail and LLM security benchmark rankings - [Prompt Security Leaderboard](https://guardion.ai/leaderboard/prompt-security): Prompt attack detection benchmark (F1-Score) - [LLM Vulnerability Leaderboard](https://guardion.ai/leaderboard/llm-vulnerability): LLM attack success rate benchmark - [Data Privacy Leaderboard](https://guardion.ai/leaderboard/data-privacy): PII detection and data protection benchmark - [Safety Moderation Leaderboard](https://guardion.ai/leaderboard/safety-moderation): Harmful content and safety moderation benchmark - [Compare Guardrails](https://guardion.ai/leaderboard/compare): Side-by-side guardrail and LLM comparisons ## AI Security Index - [AI Security Index](https://guardion.ai/ai-security-index): Comprehensive index of AI security tools and vendors - [Whitepapers](https://guardion.ai/ai-security-index/whitepapers): AI security research whitepapers - [MCP Security](https://guardion.ai/ai-security-index/mcp): MCP server security directory - [Agent Skills Security](https://guardion.ai/ai-security-index/skills): Agent skills security directory - [Alternatives](https://guardion.ai/ai-security-index/alternatives): Compare AI security vendors - [Check Point (GenAI Protect) Alternatives](https://guardion.ai/ai-security-index/alternatives/check-point) - [Lasso Security Alternatives](https://guardion.ai/ai-security-index/alternatives/lasso-security) - [Palo Alto Networks (Prisma AIRS) Alternatives](https://guardion.ai/ai-security-index/alternatives/palo-alto) - [Prompt Security (SentinelOne) Alternatives](https://guardion.ai/ai-security-index/alternatives/prompt-security) - [Zscaler (GenAI Security) Alternatives](https://guardion.ai/ai-security-index/alternatives/zscaler) - [Protect AI (Palo Alto Networks) Alternatives](https://guardion.ai/ai-security-index/alternatives/protect-ai) - [Akto Alternatives](https://guardion.ai/ai-security-index/alternatives/akto) - [Lakera (Check Point) Alternatives](https://guardion.ai/ai-security-index/alternatives/lakera) - [Mindgard Alternatives](https://guardion.ai/ai-security-index/alternatives/mindgard) - [Aim Security Alternatives](https://guardion.ai/ai-security-index/alternatives/aim-security) - [DeepKeep Alternatives](https://guardion.ai/ai-security-index/alternatives/deepkeep) - [Noma Security Alternatives](https://guardion.ai/ai-security-index/alternatives/noma) - [Cranium Alternatives](https://guardion.ai/ai-security-index/alternatives/cranium) - [Credal.ai Alternatives](https://guardion.ai/ai-security-index/alternatives/credal) - [HiddenLayer Alternatives](https://guardion.ai/ai-security-index/alternatives/hiddenlayer) - [Cisco AI Defense (Robust Intelligence) Alternatives](https://guardion.ai/ai-security-index/alternatives/cisco-robust) - [Tenable (Apex Security) Alternatives](https://guardion.ai/ai-security-index/alternatives/tenable-apex) - [Fiddler AI Alternatives](https://guardion.ai/ai-security-index/alternatives/fiddler) - [WhyLabs Alternatives](https://guardion.ai/ai-security-index/alternatives/whylabs) - [Snowflake (TruEra) Alternatives](https://guardion.ai/ai-security-index/alternatives/snowflake-truera) - [Adversa AI Alternatives](https://guardion.ai/ai-security-index/alternatives/adversa) - [Enkrypt AI Alternatives](https://guardion.ai/ai-security-index/alternatives/enkrypt) - [Private AI Alternatives](https://guardion.ai/ai-security-index/alternatives/private-ai) - [Aporia (Coralogix) Alternatives](https://guardion.ai/ai-security-index/alternatives/aporia) - [CalypsoAI (F5) Alternatives](https://guardion.ai/ai-security-index/alternatives/calypso-ai) - [TrojAI Alternatives](https://guardion.ai/ai-security-index/alternatives/trojai) - [Acuvity Alternatives](https://guardion.ai/ai-security-index/alternatives/acuvity) - [Citadel AI Alternatives](https://guardion.ai/ai-security-index/alternatives/citadel-ai) - [SPLX.ai Alternatives](https://guardion.ai/ai-security-index/alternatives/splx-ai) - [Pillar Security Alternatives](https://guardion.ai/ai-security-index/alternatives/pillar-security) - [Straiker AI Alternatives](https://guardion.ai/ai-security-index/alternatives/straiker-ai) - [Aqua Security Alternatives](https://guardion.ai/ai-security-index/alternatives/aqua-security) - [Preamble Alternatives](https://guardion.ai/ai-security-index/alternatives/preamble) - [PromptArmor Alternatives](https://guardion.ai/ai-security-index/alternatives/promptarmor) - [Securiti Alternatives](https://guardion.ai/ai-security-index/alternatives/securiti) - [Capsule Security Alternatives](https://guardion.ai/ai-security-index/alternatives/capsule-security) - [NeuralTrust Alternatives](https://guardion.ai/ai-security-index/alternatives/neuraltrust) - [Apiiro Alternatives](https://guardion.ai/ai-security-index/alternatives/apiiro) - [KELA (AiFort) Alternatives](https://guardion.ai/ai-security-index/alternatives/kela) - [Infotect Security Alternatives](https://guardion.ai/ai-security-index/alternatives/infotect) - [Keycard Alternatives](https://guardion.ai/ai-security-index/alternatives/keycard) - [Sgnl Alternatives](https://guardion.ai/ai-security-index/alternatives/sgnl) - [Galileo Alternatives](https://guardion.ai/ai-security-index/alternatives/galileo) - [Arize AI Alternatives](https://guardion.ai/ai-security-index/alternatives/arize) - [Permit.io Alternatives](https://guardion.ai/ai-security-index/alternatives/permit-io) - [Langfuse Alternatives](https://guardion.ai/ai-security-index/alternatives/langfuse) - [Harmonic Security Alternatives](https://guardion.ai/ai-security-index/alternatives/harmonic-security) - [Promptfoo Alternatives](https://guardion.ai/ai-security-index/alternatives/promptfoo) - [WitnessAI Alternatives](https://guardion.ai/ai-security-index/alternatives/witness-ai) - [Braintrust Alternatives](https://guardion.ai/ai-security-index/alternatives/braintrust) - [Patronus AI Alternatives](https://guardion.ai/ai-security-index/alternatives/patronus-ai) - [Nightfall AI Alternatives](https://guardion.ai/ai-security-index/alternatives/nightfall) - [Zenity Alternatives](https://guardion.ai/ai-security-index/alternatives/zenity) - [Archestra Alternatives](https://guardion.ai/ai-security-index/alternatives/archestra) - [Operant AI Alternatives](https://guardion.ai/ai-security-index/alternatives/operant-ai) - [Geordie AI Alternatives](https://guardion.ai/ai-security-index/alternatives/geordie-ai) - [Aurascape Alternatives](https://guardion.ai/ai-security-index/alternatives/aurascape) - [Arthur Alternatives](https://guardion.ai/ai-security-index/alternatives/arthur) - [Meta (Llama Guard) Alternatives](https://guardion.ai/ai-security-index/alternatives/llama-guard) - [NVIDIA (NeMo Guardrails) Alternatives](https://guardion.ai/ai-security-index/alternatives/nemo-guardrails) - [LLM Guard (Protect AI) Alternatives](https://guardion.ai/ai-security-index/alternatives/llm-guard) - [Giskard (Open Source) Alternatives](https://guardion.ai/ai-security-index/alternatives/giskard) - [ModelScan Alternatives](https://guardion.ai/ai-security-index/alternatives/modelscan) - [Fickling Alternatives](https://guardion.ai/ai-security-index/alternatives/fickling) - [garak (NVIDIA) Alternatives](https://guardion.ai/ai-security-index/alternatives/garak) - [Vigil Alternatives](https://guardion.ai/ai-security-index/alternatives/vigil) - [LangKit Alternatives](https://guardion.ai/ai-security-index/alternatives/langkit) - [Rebuff Alternatives](https://guardion.ai/ai-security-index/alternatives/rebuff) - [Guardrails AI Alternatives](https://guardion.ai/ai-security-index/alternatives/guardrails-ai) - [OpenGuardrails Alternatives](https://guardion.ai/ai-security-index/alternatives/openguardrails) - [Azure AI Content Safety (Prompt Shields) Alternatives](https://guardion.ai/ai-security-index/alternatives/azure-prompt-shields) - [Google Cloud Model Armor Alternatives](https://guardion.ai/ai-security-index/alternatives/google-model-armor) - [Amazon Bedrock Guardrails Alternatives](https://guardion.ai/ai-security-index/alternatives/aws-bedrock-guardrails) - [Cloudflare Firewall for AI Alternatives](https://guardion.ai/ai-security-index/alternatives/cloudflare-firewall-for-ai) - [Wiz (AI-SPM) Alternatives](https://guardion.ai/ai-security-index/alternatives/wiz) - [Okta (AI Agent Identity) Alternatives](https://guardion.ai/ai-security-index/alternatives/okta) - [Netskope One AI Security Alternatives](https://guardion.ai/ai-security-index/alternatives/netskope) - [Snyk (Invariant Labs) Alternatives](https://guardion.ai/ai-security-index/alternatives/snyk-invariant) - [Runlayer Alternatives](https://guardion.ai/ai-security-index/alternatives/runlayer) - [MintMCP Alternatives](https://guardion.ai/ai-security-index/alternatives/mintmcp) - [Formal Alternatives](https://guardion.ai/ai-security-index/alternatives/formal) - [Highflame Alternatives](https://guardion.ai/ai-security-index/alternatives/highflame) - [Oligo Security Alternatives](https://guardion.ai/ai-security-index/alternatives/oligo-security) - [Levo.ai Alternatives](https://guardion.ai/ai-security-index/alternatives/levo-ai) - [Repello AI Alternatives](https://guardion.ai/ai-security-index/alternatives/repello-ai) - [CodeIntegrity Alternatives](https://guardion.ai/ai-security-index/alternatives/codeintegrity) - [General Analysis Alternatives](https://guardion.ai/ai-security-index/alternatives/general-analysis) - [Unbound Security Alternatives](https://guardion.ai/ai-security-index/alternatives/unbound-security) - [Pangea (CrowdStrike) Alternatives](https://guardion.ai/ai-security-index/alternatives/pangea) - [Knostic Alternatives](https://guardion.ai/ai-security-index/alternatives/knostic) - [Vijil Alternatives](https://guardion.ai/ai-security-index/alternatives/vijil) - [Haize Labs Alternatives](https://guardion.ai/ai-security-index/alternatives/haize-labs) - [Virtue AI Alternatives](https://guardion.ai/ai-security-index/alternatives/virtue-ai) - [Adaptive Security Alternatives](https://guardion.ai/ai-security-index/alternatives/adaptive-security) - [Docker MCP Gateway & Catalog Alternatives](https://guardion.ai/ai-security-index/alternatives/docker-mcp-gateway) - [IBM Granite Guardian Alternatives](https://guardion.ai/ai-security-index/alternatives/ibm-granite-guardian) - [Meta LlamaFirewall Alternatives](https://guardion.ai/ai-security-index/alternatives/llamafirewall) - [Stacklok ToolHive Alternatives](https://guardion.ai/ai-security-index/alternatives/toolhive) - [mcp-scan (Snyk Agent Scan) Alternatives](https://guardion.ai/ai-security-index/alternatives/mcp-scan) ## PromptIntel Threat Database - [PromptIntel](https://guardion.ai/promptintel): AI threat intelligence database - [Threat Categories](https://guardion.ai/promptintel/categories): Browse threats by category - [Threat Prompts](https://guardion.ai/promptintel/threats/prompts): Browse malicious prompt examples - [Ansiescape](https://guardion.ai/promptintel/category/ansiescape): ANSI escape code injection attacks that disrupt terminal processing - [Apikey](https://guardion.ai/promptintel/category/apikey): Attempts to generate API keys for various services - [Atkgen](https://guardion.ai/promptintel/category/atkgen): Automated attack generation using separate models - [Audio](https://guardion.ai/promptintel/category/audio): Audio-based attacks on multimodal models - [Av Spam Scanning](https://guardion.ai/promptintel/category/av_spam_scanning): Tests for antivirus and spam scanning on outputs - [Continuation](https://guardion.ai/promptintel/category/continuation): Contentious term continuation attacks - [Dan](https://guardion.ai/promptintel/category/dan): Do-Anything-Now prompts that bypass model alignment - [Divergence](https://guardion.ai/promptintel/category/divergence): Attacks to deviate output from original topic - [Doctor](https://guardion.ai/promptintel/category/doctor): Roleplaying as doctors to override safeguards - [Donotanswer](https://guardion.ai/promptintel/category/donotanswer): Requests for contentious or unwanted content - [Dra](https://guardion.ai/promptintel/category/dra): Disguise and Reconstruction Attacks using string manipulation - [Encoding](https://guardion.ai/promptintel/category/encoding): Encoding-based injection using various character sets - [Exploitation](https://guardion.ai/promptintel/category/exploitation): Template, code, and SQL injection exploits - [Fileformats](https://guardion.ai/promptintel/category/fileformats): Vulnerabilities in file types associated with targets - [Fitd](https://guardion.ai/promptintel/category/fitd): Foot In The Door multi-turn jailbreak technique - [Glitch](https://guardion.ai/promptintel/category/glitch): Glitch token attacks causing unusual model behavior - [Goodside](https://guardion.ai/promptintel/category/goodside): Attacks discovered by Riley Goodside - [Grandma](https://guardion.ai/promptintel/category/grandma): Emotional manipulation through grandma roleplay - [Latentinjection](https://guardion.ai/promptintel/category/latentinjection): Hidden injections in legitimate contexts - [Leakreplay](https://guardion.ai/promptintel/category/leakreplay): Attacks to leak training data through replay - [Lmrc](https://guardion.ai/promptintel/category/lmrc): Language Model Risk Cards testing content safety - [Malwaregen](https://guardion.ai/promptintel/category/malwaregen): Requests to generate malware or malicious code - [Misleading](https://guardion.ai/promptintel/category/misleading): Poses false claims to test model rejection - [Packagehallucination](https://guardion.ai/promptintel/category/packagehallucination): Elicits non-existent package names in code - [Phrasing](https://guardion.ai/promptintel/category/phrasing): Rephrasing requests to bypass safeguards - [Promptinject](https://guardion.ai/promptintel/category/promptinject): Basic prompt injection techniques - [Realtoxicityprompts](https://guardion.ai/promptintel/category/realtoxicityprompts): Prompts with high toxicity elicitation rates - [Sata](https://guardion.ai/promptintel/category/sata): Simple Assistive Task Linkage attacks - [Smuggling](https://guardion.ai/promptintel/category/smuggling): Token smuggling to bypass filters - [Snowball](https://guardion.ai/promptintel/category/snowball): Complex reasoning questions causing wrong answers - [Suffix](https://guardion.ai/promptintel/category/suffix): Adversarial suffix-based system prompt disruption - [Tap](https://guardion.ai/promptintel/category/tap): Tree of Attacks with Pruning jailbreak method - [Topic](https://guardion.ai/promptintel/category/topic): Attempts to engage on blocked topics via lexical space - [Visual Jailbreak](https://guardion.ai/promptintel/category/visual_jailbreak): Visual instructions to bypass text alignment - [Web Injection](https://guardion.ai/promptintel/category/web_injection): Web injection attacks like CSRF and XSS ## Risk & Incidents - [AI Incidents](https://guardion.ai/ai-incidents): Database of real-world AI security incidents - [Risk Assessment](https://guardion.ai/risk-assessment): Evaluate your AI security posture ## Blog - [The AI Agent Quarantine Field Guide: Automating Incident Response with SIEM-Integrated Gateways](https://guardion.ai/blog/ai-agent-quarantine-incident-response-playbook): Learn how to integrate AI Gateways with SIEMs to build automated incident response playbooks and quarantine rogue AI agents in production. - [EU AI Act Articles 12 & 14 Decoded: Logging and Human Oversight for AI Agents](https://guardion.ai/blog/eu-ai-act-compliance-ai-agents-article-12-14): Learn how to secure autonomous AI agents for EU AI Act compliance. Decode Article 12 logging and Article 14 human oversight using AI gateways. - [Phantom Workflows: Why Traditional EDR Fails for Ephemeral AI Agents](https://guardion.ai/blog/ephemeral-ai-agents-edr-runtime-security): Discover why legacy EDR and static LLM firewalls fail to secure ephemeral AI agents, and how Agentic EDR protects your autonomous workflows at runtime. - [Benchmarking Agent DLP: Why LLM-as-a-Judge Fails at Real-Time PII Redaction](https://guardion.ai/blog/llm-pii-redaction-latency-benchmark-ai-agents): Discover why LLM-as-a-judge is too slow for real-time AI data loss prevention. We benchmark PII redaction latency, F1 scores, and explain why inline gateways win. - [The Agent FinOps Field Guide: Instrumenting Gateways to Track Token Usage and Execution Costs](https://guardion.ai/blog/agent-finops-field-guide-gateway-token-cost-tracking): Learn how to implement Agentic FinOps using an inline AI gateway to track token usage, intercept MCP costs, and build enterprise observability without SDKs. - [The MCP SecOps Field Guide: Detecting and Blocking Tool Poisoning Attacks](https://guardion.ai/blog/mcp-security-tool-poisoning-field-guide): Learn how to detect and block MCP Tool Poisoning Attacks (TPAs). A complete SecOps field guide to securing the Model Context Protocol against data exfiltration. - [The OWASP LLM01 Field Guide: Governing AI Agents Against Role Confusion and In-the-Wild IDPI](https://guardion.ai/blog/owasp-llm01-field-guide-indirect-prompt-injection-role-confusion): Master OWASP LLM01 with this field guide. Learn how to defend AI agents against in-the-wild indirect prompt injection, Role Confusion, and CoT Forgery. - [NIST AI Agent Security Decoded: How to Govern Autonomous Workflows](https://guardion.ai/blog/nist-ai-agent-security-decoded-how-to-govern-autonomous-workflows): Master the NIST AI agent security framework. Learn how to govern autonomous systems, secure MCP tool calls, and align your AI agents with NIST standards. - [MITRE ATLAS Decoded: Mapping the Attack Lifecycle of AI Agents](https://guardion.ai/blog/mitre-atlas-decoded-ai-agent-attack-lifecycle): Learn how to map and defend against AI agent attacks using the MITRE ATLAS framework. Discover the agent kill chain and how to secure the AI runtime. - [Meta's Agents Rule of Two Decoded: How to Defend Against the Lethal Trifecta](https://guardion.ai/blog/meta-agents-rule-of-two-decoded-lethal-trifecta): Learn how to implement Meta's "Agents Rule of Two" to secure your autonomous workflows against the lethal trifecta of AI prompt injection. - [ISO 42001 Decoded: Governing Autonomous AI Agents in Fintech and Healthcare](https://guardion.ai/blog/iso-42001-compliance-ai-agents): Learn how to implement ISO 42001 controls for autonomous AI agents. A CISO's field guide to governing agentic workflows in Fintech and Healthcare. - [DORA Compliance Decoded: Securing Financial AI Agents for Operational Resilience](https://guardion.ai/blog/dora-compliance-decoded-securing-financial-ai-agents-for-operational-resilience): Master DORA compliance for AI agents. Learn how to secure autonomous financial workflows, govern MCP tool calls, and ensure operational resilience. - [Zero-Click AI Agent Attacks: How Untrusted Content Becomes Remote Code Execution](https://guardion.ai/blog/zero-click-ai-agent-attacks): Cataloging the 0-click attack pattern across AI agents—how untrusted content triggers autonomous actions leading to data theft or RCE, and how to defend against it at the gateway layer. - [Your AI IDE Is an Attack Surface: Securing Cursor, Copilot, and Claude Code](https://guardion.ai/blog/your-ai-ide-is-an-attack-surface-securing-cursor-copilot-and-claude-code): AI-powered IDEs like Cursor, GitHub Copilot, and Claude Code introduce new supply chain and execution risks. Learn how to secure your agentic development environment. - [Why AI Safety Classifiers Fail: Training Overfit, Decision Boundaries, and Blind Spots](https://guardion.ai/blog/why-ai-safety-classifiers-fail): Explore the technical reasons why AI safety classifiers fail in production, from training overfit to adversarial blind spots, and learn how attackers bypass LLM guardrails. - [Tool-Use Monitoring for AI Agents: Detecting Over-Privileged Actions Before They Execute](https://guardion.ai/blog/tool-use-monitoring-for-ai-agents-detecting-over-privileged-actions-before-they-execute): Monitor and control AI agent tool usage with gateway-level authorization. Prevent over-privileged actions, secure MCP servers, and audit agent behavior. - [The MCP Security Crisis: Defending Against Indirect Prompt Injection in the Model Context Protocol](https://guardion.ai/blog/the-mcp-security-crisis-defending-against-indirect-prompt-injection-in-the-model-context-protocol): Learn how the Model Context Protocol (MCP) exposes AI agents to indirect prompt injection and tool poisoning, and how to secure your integrations using an AI Gateway. - [Sub-Millisecond AI Security: Building Guardrails That Don't Slow Down Your Agents](https://guardion.ai/blog/sub-millisecond-ai-security): Learn how to build AI guardrails that add less than 1ms of latency. Benchmarks, edge deployment patterns, and streaming-compatible security techniques. - [Shadow AI in the Enterprise: How to Detect and Secure Unauthorized LLM Usage Through Your Gateway](https://guardion.ai/blog/shadow-ai-in-the-enterprise): Discover how to detect, monitor, and secure unauthorized LLM and AI agent usage across your enterprise using a network-level AI Security Gateway. - [Securing the AI-Augmented Developer Workflow: From Code Generation to Deployment](https://guardion.ai/blog/securing-the-ai-augmented-developer-workflow-from-code-generation-to-deployment): A comprehensive guide to securing the AI-augmented developer workflow, mitigating risks in code generation, and protecting AI agents across the SDLC. - [Securing Multimodal AI Pipelines: Gateway-Level Inspection for Images, Audio, and Documents](https://guardion.ai/blog/securing-multimodal-ai-pipelines): Explore the expanding attack surface of multimodal LLMs and learn how gateway-level inspection protects AI pipelines against visual prompt injection, malicious documents, and audio threats. - [Securing Microsoft Copilot Studio: Attack Vectors and Gateway-Level Defenses](https://guardion.ai/blog/securing-microsoft-copilot-studio-attack-vectors-and-gateway-level-defenses): Explore emerging attack vectors targeting Microsoft Copilot Studio, including AIjacking and data exfiltration, and learn how gateway-level defenses provide essential enterprise protection. - [Securing AI Plugin Ecosystems: Lessons From OpenClaw, GPT Store, and MCP Marketplaces](https://guardion.ai/blog/securing-ai-plugin-ecosystems): As AI agents increasingly rely on third-party plugins and MCP tools, the attack surface expands exponentially. Learn how to secure your AI ecosystem against prompt injection, tool poisoning, and malicious execution. - [Salesforce Einstein Security: Preventing Prompt Mines and Data Corruption](https://guardion.ai/blog/salesforce-einstein-security-preventing-prompt-mines-and-data-corruption): Discover how prompt mines cause 0-click data corruption in Salesforce Einstein and learn how to secure your CRM's AI agents against novel injection threats. - [Red Teaming Your AI Gateway: A Practical Checklist for Guardrail Validation](https://guardion.ai/blog/red-teaming-your-ai-gateway-a-practical-checklist-for-guardrail-validation): A comprehensive checklist and methodology for red teaming AI gateways, validating guardrails against prompt injections, tool poisoning, and data exfiltration. - [Multimodal Prompt Injection: How Images, Audio, and Video Bypass Text-Only Defenses](https://guardion.ai/blog/multimodal-prompt-injection-how-images-audio-and-video-bypass-text-only-defenses): Discover how multimodal prompt injection uses images, audio, and video to bypass text-only defenses in LLMs, and learn how to protect your AI agents. - [Multi-Layer Defense: Why No Single Classifier Can Protect Your LLM Application](https://guardion.ai/blog/multi-layer-defense-why-no-single-classifier-can-protect-your-llm-application): Discover why a defense in depth AI strategy is essential. Learn how a multi-layer AI defense architecture protects LLM applications when single classifiers fail. - [MITRE ATLAS for AI Agents: Mapping Real Attack Techniques to Your Defense Strategy](https://guardion.ai/blog/mitre-atlas-for-ai-agents): Learn how to use the MITRE ATLAS framework to model threats and defend against real-world attack techniques on AI agents and MCP servers. - [Memory Poisoning in AI Agents: How Persistent Context Becomes a Persistent Threat](https://guardion.ai/blog/memory-poisoning-in-ai-agents): Explore how AI agent memory poisoning and persistent context manipulation lead to instruction drift, and learn how to secure your AI memory from malicious attacks. - [How to Build an AI Red Team Program: From Ad Hoc Testing to Continuous Evaluation](https://guardion.ai/blog/how-to-build-an-ai-red-team-program): A comprehensive guide to building a continuous AI red teaming program to secure LLMs and AI agents against adversarial attacks, moving beyond ad hoc testing. - [How Jira Tickets, Pull Requests, and Docs Become AI Exploit Vectors](https://guardion.ai/blog/how-jira-tickets-pull-requests-and-docs-become-ai-exploit-vectors): Discover how AI agents reading Jira tickets, GitHub PRs, and internal docs can be exploited via indirect prompt injection, and how to secure developer workflows. - [Guardrails Are Not Enough: Why AI Security Needs Hard Boundaries and Runtime Enforcement](https://guardion.ai/blog/guardrails-are-not-enough-why-ai-security-needs-hard-boundaries-and-runtime-enforcement): Why AI guardrails alone are not enough to protect your applications. Learn the difference between guardrails and hard boundaries, and why runtime AI enforcement is critical for secure AI architecture. - [Agent Security & Autonomous AI Risks: Why Guardrails Aren't Enough](https://guardion.ai/blog/group-2-agent-security-and-autonomous-ai-risks): A deep dive into securing multi-agent systems, tool-use monitoring, and preventing data exfiltration using gateway-level controls rather than post-hoc detection. - [Guardrails vs. Hard Boundaries: The Philosophy of AI Security](https://guardion.ai/blog/group-14-guardrails-vs-hard-boundaries-the-philosophy-of-ai-security): Why probabilistic guardrails aren't enough for AI agents, and how network-level hard boundaries deliver verifiable runtime security. - [AI Observability, Monitoring & Incident Response: A Definitive Guide for Agentic Systems](https://guardion.ai/blog/group-12-ai-observability-monitoring-incident-response): How to monitor AI agents in production, detect anomalies with real-time metrics, and execute an incident response playbook from detection to containment. - [Data-Structure Injection & Novel AI Attack Vectors](https://guardion.ai/blog/group-10-data-structure-injection-novel-ai-attack-vectors): Prompt injection was just the beginning. Discover how attackers are exploiting data structures, poisoning memory, and compromising AI agents through novel attack vectors. - [AI Gateway Security & Runtime Protection](https://guardion.ai/blog/group-1-ai-gateway-security-runtime-protection): How to position an AI Gateway as the security control plane for agentic traffic, intercepting threats and redacting PII before they reach your LLM. - [Defending Against AIjacking: Runtime Protection for Enterprise AI Assistants](https://guardion.ai/blog/defending-against-aijacking): Learn how to defend against AIjacking, a critical attack category targeting enterprise AI assistants, using gateway-level runtime protection. - [Cross-Platform AI Security: One Gateway for AWS Bedrock, Azure AI, and Google Vertex](https://guardion.ai/blog/cross-platform-ai-security-one-gateway-for-aws-bedrock-azure-ai-and-google-vertex): Securing multi-cloud AI deployments requires a unified approach. Learn why relying on vendor-specific AI security creates dangerous blind spots and how a centralized AI Gateway solves this. - [Building an AI Security Dashboard: Real-Time Metrics That Actually Matter](https://guardion.ai/blog/building-an-ai-security-dashboard-real-time-metrics-that-actually-matter): How to build an AI security dashboard that cuts through the noise. Learn which metrics matter for LLM monitoring, AI gateways, and real-time agent observability. - [Beyond Regex: Detecting Prompt Injection at the Gateway Layer Using ML Classification](https://guardion.ai/blog/beyond-regex-detecting-prompt-injection-at-the-gateway-layer-using-ml-classification): Move beyond regex and LLM-as-judge for prompt injection detection. Learn how ML classifiers provide sub-millisecond, accurate prompt injection defense at the gateway. - [Beyond Prompt Injection: 5 Emerging AI Attack Vectors You Need to Know in 2026](https://guardion.ai/blog/beyond-prompt-injection-5-emerging-ai-attack-vectors-2026): Discover the top 5 emerging AI attack vectors in 2026 beyond prompt injection, including MCP Tool Poisoning and RAG Context Exfiltration, and how to protect agentic systems. - [Automated AI Red Teaming: Using Adversarial Agents to Test Your Defenses at Scale](https://guardion.ai/blog/automated-ai-red-teaming-using-adversarial-agents-to-test-your-defenses-at-scale): Manual LLM penetration testing can't keep up with weekly agent deployments. Learn how to use agent-to-agent exploitation for continuous, automated AI security testing. - [GuardionAI (Metatext) Selected for ANPD AI Regulatory Sandbox](https://guardion.ai/blog/anpd-blog-post): Metatext, the company behind GuardionAI, has been selected in first place for Brazil's ANPD AI Regulatory Sandbox to pioneer responsible AI innovation and data protection. - [AI SSRF Attacks: How Malicious MCP Servers Expose Your Internal Network (And How to Stop It)](https://guardion.ai/blog/ai-ssrf-attacks-mcp-prompt-injection): Learn how AI agents and malicious MCP servers expose your internal networks to SSRF attacks, and how to stop them using an AI Security Gateway. - [The AI Security Middleware Stack: PII, Prompt Injection & Content Moderation in One Pipeline](https://guardion.ai/blog/ai-security-middleware-stack): Build a unified AI security pipeline for PII detection, prompt injection, and content moderation. One middleware stack, one request lifecycle. - [AI Security in 2026: What Changed, What Didn't, and What's Coming Next](https://guardion.ai/blog/ai-security-in-2026-what-changed-what-didnt-and-whats-coming-next): An analysis of the AI threat landscape in 2026, comparing predictions from 2025 to reality, and how agentic systems have shifted the security paradigm. - [AI Incident Response Playbook: From Detection to Containment in Agentic Systems](https://guardion.ai/blog/ai-incident-response-playbook): A comprehensive playbook for handling AI security incidents. Learn how to detect, contain, and recover from rogue agents, prompt injections, and tool poisoning. - [AI Security Compliance Checklist 2026: EU AI Act, NIST AI RMF, and OWASP LLM Top 10 in Practice](https://guardion.ai/blog/ai-compliance-checklist-2026): The complete 2026 AI compliance checklist mapping EU AI Act, NIST AI RMF, and OWASP LLM Top 10 to concrete technical controls you can implement today. - [Voice AI Security: Preventing Adversarial Audio Attacks on AI Assistants](https://guardion.ai/blog/voice-ai-security-preventing-adversarial-audio-attacks): Discover how adversarial audio attacks manipulate voice AI assistants, and learn actionable mitigation strategies to secure your agentic systems. - [AI Security Gateway: How Proxying LLM and MCP Calls Enables Security Observability, PII Redaction, and Policy Enforcement](https://guardion.ai/blog/ai-security-gateway-how-proxying-llm-and-mcp-calls-enables-security-observability-pii-redaction-and-policy-enforcement): Discover how an AI Security Gateway works as a transparent proxy to intercept LLM and MCP calls, providing critical security observability, PII redaction, and policy enforcement without code changes. - [Building Adversarial-Resistant Guardrails: Lessons From Red Teaming Production Classifiers](https://guardion.ai/blog/building-adversarial-resistant-guardrails): Learn how to build robust, adversarial-resistant AI guardrails by understanding common bypass techniques and red teaming strategies for production LLM classifiers. - [AI Agent Supply Chain Attacks: From Malicious Skills to Compromised MCP Servers](https://guardion.ai/blog/ai-agent-supply-chain-attacks): Explore the emerging threats in the AI agent supply chain, focusing on malicious skills and compromised MCP servers, and learn how to defend your infrastructure. - [Data Exfiltration via AI Agents: Attack Patterns and Gateway-Level Defenses](https://guardion.ai/blog/ai-agent-data-exfiltration-defense): Learn 7 techniques attackers use to exfiltrate data through AI agents and how gateway-level defenses prevent them. PII detection, output filtering, and more. - [OWASP AI Exchange: The World's Definitive AI Security Guide (2026)](https://guardion.ai/blog/owasp-ai-exchange-worlds-definitive-ai-security-guide-2026): Comprehensive guide to the OWASP AI Exchange - an open-source flagship project providing 300+ pages of practical AI security guidance. Learn how this authoritative resource shapes ISO/IEC 27090, EU AI Act compliance, and serves as the go-to framework for practitioners securing generative and predictive AI systems. - [LLM Compliance 2026: Complete Guide to ISO 42001, EU AI Act, SOC 2, and GDPR](https://guardion.ai/blog/llm-compliance-guide-iso-42001-eu-ai-act-soc2-gdpr-2026): Comprehensive guide to navigating AI compliance frameworks for production LLM systems. Covers ISO 42001 certification, EU AI Act risk categories, SOC 2 audit requirements, and GDPR data protection with practical implementation examples, code snippets, and 90-day compliance roadmap. - [Indirect Prompt Injection in RAG Systems: How Your Knowledge Base Became an Attack Vector](https://guardion.ai/blog/rag-security-indirect-prompt-injection): Secure your RAG pipeline from indirect prompt injection. Learn how attackers poison knowledge bases and how gateway-level document scanning prevents it. - [AI Gateway Security: How to Intercept and Neutralize Threats Before They Reach Your LLM](https://guardion.ai/blog/ai-gateway-security): Learn how AI gateways intercept prompt injection, data leaks, and abuse before they reach your LLM. Architecture patterns and benchmarks included. - [GuardionAI Selected for ANPD's AI Regulatory Sandbox: Pioneering Responsible AI Innovation in Brazil](https://guardion.ai/blog/metatext-selected-anpd-ai-regulatory-sandbox): GuardionAI (Metatext) has been selected as one of three companies to participate in Brazil's ANPD AI Regulatory Sandbox. This groundbreaking initiative will run until December 2026, focusing on algorithmic transparency and personal data protection while advancing responsible AI technology development. - [GuardionAI Selected for Google for Startups Accelerator: AI for Cybersecurity](https://guardion.ai/blog/guardionai-google-accelerator-ai-cybersecurity-latam): GuardionAI has been selected as one of 11 startups to join Google for Startups Accelerator: AI for Cybersecurity. This prestigious 10-week program brings together the best of Google's technology, mentorship, and resources to support startups leveraging AI to address complex cybersecurity threats. - [Runtime Guardrails Index: GuardionAI Prompt Defense vs Azure PromptShield, AWS Guardrails, and GCP Model Armor](https://guardion.ai/blog/benchmark-azure-prompt-shield-aws-guardrails-gcp-model-armor): Comprehensive benchmark comparing GuardionAI Prompt Defense to Azure Prompt Shield, AWS Guardrails, and GCP Model Armor across multilingual prompt injection and adversarial attacks. GuardionAI stands out with industry-leading F1 score, recall, and lowest latency, demonstrating superior protection and performance over major cloud providers. - [ModernGuard: The Multilingual and Ultra-Fast Prompt Attack Detector for Agent Security](https://guardion.ai/blog/modern-guard-multilingual-ultra-fast-prompt-injection-guardrail): GuardionAI introduces ModernGuard, a multilingual, ultra-fast transformer-encoder for LLM runtime guardrail / threat detection, trained on millions of prompt attack simulations, red teaming and threat databases (public and private). ## Legal - [Privacy Policy](https://guardion.ai/privacy-policy) - [Terms of Service](https://guardion.ai/terms-of-service) - [Cookie Policy](https://guardion.ai/cookie-policy) - [Data Processing Agreement](https://guardion.ai/data-processing-agreement) ## Sitemaps - [Sitemap Index](https://guardion.ai/sitemap.xml): Root directory of all section sitemaps below - [Core Pages](https://guardion.ai/sitemap-core.xml): Static marketing pages (home, pricing, product, solutions, legal) - [Blog](https://guardion.ai/blog/sitemap.xml): Blog hub and all posts - [Docs](https://guardion.ai/docs/sitemap.xml): Documentation, guides and API reference - [AI Security Index Directory](https://guardion.ai/ai-security-index/sitemap-directory.xml): Curated index pages and whitepapers - [AI Security Index Best-of](https://guardion.ai/ai-security-index/sitemap-best.xml): Best-of tool roundup pages - [PromptIntel](https://guardion.ai/promptintel/sitemap.xml): Threat intelligence categories and threats - [AI Incidents](https://guardion.ai/ai-incidents/sitemap.xml): Curated AI incidents database - [Leaderboard](https://guardion.ai/leaderboard/sitemap.xml): Leaderboard hubs, model pages and comparisons - [AI Security Index](https://guardion.ai/ai-security-index/sitemap.xml): Section-level index for the two AI Security Index sitemaps