# Wardstone

> Wardstone is an LLM security API and AI guardrail that detects prompt injections, jailbreaks, harmful content, data leakage, and unknown links in a single inference call with ~30ms latency. It acts as an LLM firewall, providing bidirectional protection for AI applications by scanning both user inputs and model outputs.

Wardstone Guard is a unified LLM security model for comprehensive AI content moderation and prompt defense. The API accepts text input and returns `flagged`, per-category `risk_bands`, and `primary_category` for four threat types. Paid plans can optionally include raw model scores with `include_raw_scores=true`.

- **Prompt Attacks**: Jailbreaks, prompt injections, and attempts to bypass system instructions
- **Harmful Content**: Violence, hate speech, sexual content, self-harm, criminal activity
- **Data Leakage**: PII exposure including SSNs, credit cards, phone numbers, emails, addresses
- **Unknown Links**: Suspicious or unverified URLs not on the allowlist

The service is model-agnostic and works with any LLM provider including OpenAI, Anthropic, Google, Meta, Mistral, and others.

## Quick Start

Essential resources for integrating Wardstone:

- [API Docs](https://wardstone.ai/docs): Interactive API documentation with sidebar navigation, code examples in 8 languages, response explorer, and embedded playground
- [Playground](https://wardstone.ai/playground): Interactive demo to test the Guard API
- [Dashboard Playground](https://wardstone.ai/dashboard/playground): Test the Guard API with your subscription quota (login required)
- [Detection API](https://wardstone.ai/api/detect): POST endpoint for threat detection
- [Integrations](https://wardstone.ai/integrations): Setup guides for 16+ LLM providers

## Pages

- [Home](https://wardstone.ai/): Overview of Wardstone's AI security capabilities and pricing
- [API Docs](https://wardstone.ai/docs): Interactive API docs with sidebar nav, scroll-spy, tabbed code examples, response explorer, and embedded playground
- [Blog](https://wardstone.ai/blog): AI security insights, guides, and best practices
- [Threats](https://wardstone.ai/threats): LLM security threat encyclopedia with attack vectors
- [Jailbreaks](https://wardstone.ai/jailbreaks): LLM jailbreak detection guides for ChatGPT, Claude, Gemini, and more
- [Compare](https://wardstone.ai/compare): Comparison of Wardstone vs competitor solutions
- [Solutions](https://wardstone.ai/solutions): Industry-specific AI security solutions
- [Pricing](https://wardstone.ai/pricing): Plans and pricing for Wardstone AI security API
- [Enterprise](https://wardstone.ai/enterprise): Enterprise AI security solutions with custom pricing and on-premise deployment
- [Contact Sales](https://wardstone.ai/contact): Enterprise contact form
- [Login](https://wardstone.ai/login): Sign in and create an account
- [Forgot Password](https://wardstone.ai/forgot-password): Request a password reset link
- [Reset Password](https://wardstone.ai/reset-password): Set a new password with a reset token
- [Verify Email](https://wardstone.ai/verify-email): Verify a new account email address
- [Privacy Policy](https://wardstone.ai/privacy): Privacy policy and data practices
- [Terms of Service](https://wardstone.ai/terms): Terms and conditions for using our services

## Blog Articles

Expert guides and best practices for AI security:

- [AI Security Trends and Predictions for 2026](https://wardstone.ai/blog/ai-security-trends-2026): The top AI security trends shaping 2026, from agentic AI risks to EU AI Act enforcement. Data-backed predictions for CTOs and security leaders.
- [Why Input Validation Alone Can't Secure Your LLM Application](https://wardstone.ai/blog/input-validation-not-enough-llm-security): Discover why LLM input validation with regex and blocklists fails against prompt injection, and how ML-based detection provides the security layers your AI app needs.
- [The Developer's Guide to Responsible AI Deployment](https://wardstone.ai/blog/responsible-ai-deployment-guide): A practical guide to responsible AI deployment for developers. Pre-launch checklists, bias testing, safety guardrails, and ethical considerations for shipping AI features.
- [AI Security Monitoring: The Metrics You Should Be Tracking](https://wardstone.ai/blog/ai-security-monitoring-metrics): Learn which AI security monitoring metrics matter most for LLM apps. Covers detection rates, latency, false positives, alerting thresholds, and dashboards.
- [Defense-in-Depth for LLM Applications: A Layered Security Approach](https://wardstone.ai/blog/defense-in-depth-llm-applications): Learn how to apply defense-in-depth principles to LLM applications with layered security controls covering input validation, guardrails, and monitoring.
- [How Adversarial Attacks Work Against Language Models](https://wardstone.ai/blog/adversarial-attacks-language-models): Learn how adversarial attacks exploit LLM tokenization, embeddings, and alignment. Covers GCG, AutoDAN, token manipulation, and proven defense strategies.
- [AI Security in Financial Services: Navigating Compliance and Risk](https://wardstone.ai/blog/ai-security-financial-services): How financial services teams can secure AI deployments while meeting SEC, OCC, and EU AI Act compliance requirements for LLM security.
- [Fine-Tuning vs Guardrails: Two Approaches to LLM Safety](https://wardstone.ai/blog/fine-tuning-vs-guardrails-llm-safety): Compare fine-tuning (RLHF, DPO, constitutional AI) with external guardrails for LLM safety. Learn why production systems need both approaches working together.
- [The Cost of Insecure AI: Calculating Risk for Your Organization](https://wardstone.ai/blog/cost-of-insecure-ai): Learn the true cost of AI security breaches and how to calculate risk for your organization. Includes real breach data, ROI frameworks, and actionable steps.
- [AI Security Incident Response: A Runbook for Engineering Teams](https://wardstone.ai/blog/ai-security-incident-response): A practical AI security incident response runbook for engineering teams. Step-by-step playbook covering detection, containment, and recovery for LLM breaches.
- [Building a Content Moderation Pipeline for AI Applications](https://wardstone.ai/blog/content-moderation-pipeline-ai): Learn how to build a production-ready AI content moderation pipeline. Covers architecture, pre and post-processing, scaling strategies, and code examples.
- [Prompt Injection vs Jailbreaking: What's the Difference?](https://wardstone.ai/blog/prompt-injection-vs-jailbreaking): Learn the key differences between prompt injection and jailbreaking in LLMs. Understand attack types, real examples, and why the distinction matters for defense.
- [AI Compliance in Healthcare: What HIPAA Means for LLM Applications](https://wardstone.ai/blog/hipaa-ai-compliance-healthcare): Learn what HIPAA AI compliance means for healthcare LLM applications, from PHI protection and BAA requirements to a practical deployment checklist.
- [Securing AI Agents: Why Chatbot Security Isn't Enough](https://wardstone.ai/blog/securing-ai-agents): AI agents face unique security risks that chatbot defenses can't handle. Learn about agent-specific attack vectors, threat models, and how to protect agentic AI systems.
- [Multi-Modal Attacks on AI Systems: Images, Audio, and Beyond](https://wardstone.ai/blog/multimodal-ai-attacks): Explore how multimodal AI attacks use images, audio, and cross-modal vectors to bypass LLM safety filters, plus practical defense strategies.
- [Evaluating AI Safety Tools: Benchmarks That Actually Matter](https://wardstone.ai/blog/ai-safety-benchmarks-evaluation): A practical guide to AI safety benchmarks for evaluating LLM security tools. Learn which metrics matter, what benchmarks miss, and how to test effectively.
- [How to Implement AI Guardrails Without Killing UX](https://wardstone.ai/blog/ai-guardrails-without-killing-ux): Learn how to implement AI guardrails that protect users without degrading experience. Covers latency budgets, async architectures, and UX patterns.
- [Understanding Indirect Prompt Injection: The Hidden Attack Vector](https://wardstone.ai/blog/indirect-prompt-injection-attacks): Learn how indirect prompt injection attacks work, why they bypass traditional defenses, and how to protect your AI systems from this hidden threat vector.
- [AI Security for Startups: A Practical Playbook](https://wardstone.ai/blog/ai-security-startups-playbook): A practical AI security playbook for startups. Budget-friendly strategies to protect your LLM features from prompt injection, data leakage, and attacks.
- [Data Leakage in LLMs: How PII Escapes Your Models](https://wardstone.ai/blog/pii-data-leakage-llms): Learn how LLMs leak personally identifiable information through training data extraction, context windows, and RAG pipelines, and how to prevent it.
- [Building Secure RAG Pipelines: A Developer's Guide](https://wardstone.ai/blog/secure-rag-pipelines): Learn how to build secure RAG pipelines that defend against indirect prompt injection, data poisoning, and leakage across every stage.
- [What Are AI Guardrails? A Complete Guide for Developers](https://wardstone.ai/blog/ai-guardrails): Learn what AI guardrails are, the different types, and how to implement them in production. Covers input guards, output filters, and architectural patterns.
- [AI Content Moderation: Moving Beyond Keyword Filtering](https://wardstone.ai/blog/ai-content-moderation-beyond-keywords): Learn why keyword filtering fails for AI content moderation and how ML-based approaches like multi-label classification deliver better accuracy with fewer false positives.
- [LLM Safety: Risks, Categories, and How to Mitigate Them](https://wardstone.ai/blog/llm-safety): A complete guide to LLM safety covering the main risk categories, real-world failures, and practical mitigation strategies for production AI applications.
- [The OWASP Top 10 for LLM Applications Explained](https://wardstone.ai/blog/owasp-top-10-llm-explained): A breakdown of the OWASP Top 10 for LLM Applications (2025), covering each vulnerability with real-world examples and practical mitigation strategies.
- [What Is an LLM Firewall? Architecture and Deployment Patterns](https://wardstone.ai/blog/llm-firewall): Learn what an LLM firewall is, how it differs from traditional firewalls, and how to deploy one to protect your AI applications from prompt attacks and data leakage.
- [What Is an LLM Guard? How Real-Time Detection Protects AI Apps](https://wardstone.ai/blog/llm-guard): Learn what an LLM guard is, how it works, and why every production AI app needs one. Covers detection categories, architecture, and implementation.
- [How to Detect Prompt Injection Attacks in Production](https://wardstone.ai/blog/detect-prompt-injection-production): Learn how to detect prompt injection attacks in production LLM applications. Covers rule-based, ML-based, and hybrid detection with code examples.
- [What is LLM Red Teaming and Why It Matters](https://wardstone.ai/blog/llm-red-teaming-guide): Learn what LLM red teaming is, how it works, and why it's essential for securing AI applications. Covers tools, frameworks, and practical methodologies.
- [How to Build an AI Security Program: A CTO's Guide](https://wardstone.ai/blog/building-ai-security-program): Learn how to build an effective AI security program for your organization. Step-by-step guide for CTOs covering risk assessment, tooling, and governance.
- [The Complete Guide to Prompt Injection Prevention in 2026](https://wardstone.ai/blog/prompt-injection-prevention-guide): Learn how to protect your AI apps from prompt injection attacks. Covers detection techniques, prevention strategies, and code examples for 2026.
- [LLM Security Best Practices: A Developer's Checklist](https://wardstone.ai/blog/llm-security-best-practices): A practical LLM security checklist for developers. Protect your AI applications from prompt injection, data leakage, and other vulnerabilities.

**For AI agents**: Request any page with `Accept: text/markdown` header for clean Markdown (~30x fewer tokens).

## Threat Library

Individual threat pages with detailed information, examples, and prevention:

- [Prompt Injection](https://wardstone.ai/threats/prompt-injection): Attacks that embed malicious instructions in user input
- [Jailbreak Attacks](https://wardstone.ai/threats/jailbreak-attacks): Techniques to bypass LLM safety guidelines
- [Indirect Prompt Injection](https://wardstone.ai/threats/indirect-prompt-injection): Attacks via external data sources
- [Data Leakage](https://wardstone.ai/threats/data-leakage): Unintended exposure of sensitive training data
- [PII Exposure](https://wardstone.ai/threats/pii-exposure): Leakage of personally identifiable information
- [Adversarial Prompts](https://wardstone.ai/threats/adversarial-prompts): Crafted inputs to produce incorrect outputs
- [System Prompt Extraction](https://wardstone.ai/threats/system-prompt-extraction): Attempts to reveal hidden system prompts
- [Toxic Content Generation](https://wardstone.ai/threats/toxic-content-generation): Eliciting harmful or offensive outputs
- [Hallucination Attacks](https://wardstone.ai/threats/hallucination-attacks): Exploiting model confabulation
- [Model Extraction](https://wardstone.ai/threats/model-extraction): Attempts to steal model capabilities
- [Training Data Extraction](https://wardstone.ai/threats/training-data-extraction): Recovering training data from model
- [Prompt Leaking](https://wardstone.ai/threats/prompt-leaking): Exposing conversation context
- [Social Engineering LLM](https://wardstone.ai/threats/social-engineering-llm): Manipulating AI through psychological tactics
- [Context Manipulation](https://wardstone.ai/threats/context-manipulation): Tampering with conversation state
- [DoS Attacks on LLM](https://wardstone.ai/threats/dos-attacks-llm): Resource exhaustion attacks

## LLM Jailbreak Detection Guides

Detection and prevention guides for jailbreak attacks organized by AI lab:

### OpenAI
- [ChatGPT Jailbreak Prompts](https://wardstone.ai/jailbreaks/chatgpt-jailbreak-prompts): Detection and prevention of ChatGPT jailbreak attacks
- [ChatGPT DAN Jailbreak](https://wardstone.ai/jailbreaks/chatgpt-dan-jailbreak): Detect and block DAN (Do Anything Now) attacks
- [ChatGPT Prompt Injection](https://wardstone.ai/jailbreaks/chatgpt-prompt-injection): Defend against prompt injection in ChatGPT apps
- [ChatGPT Developer Mode](https://wardstone.ai/jailbreaks/chatgpt-developer-mode): Detect developer mode jailbreak attempts
- [ChatGPT System Prompt Extraction](https://wardstone.ai/jailbreaks/chatgpt-system-prompt-extraction): Block system prompt extraction attacks
- [GPT-5 Jailbreak](https://wardstone.ai/jailbreaks/gpt-5-jailbreak): Detect and block attacks on GPT-5.2 and GPT-5.3-Codex

### Anthropic
- [Claude Jailbreak Prompts](https://wardstone.ai/jailbreaks/claude-jailbreak-prompts): Detection and prevention for Anthropic Claude
- [Claude Opus 4.6 Jailbreak](https://wardstone.ai/jailbreaks/claude-opus-4-6-jailbreak): Detection and prevention for Claude Opus 4.6
- [Claude Opus 4.5 Jailbreak](https://wardstone.ai/jailbreaks/claude-opus-4-5-jailbreak): Detection and prevention for Claude Opus 4.5
- [Claude Sonnet 4.5 Jailbreak](https://wardstone.ai/jailbreaks/claude-sonnet-4-5-jailbreak): Detection and prevention for Claude Sonnet 4.5

### Google
- [Gemini Jailbreak Prompts](https://wardstone.ai/jailbreaks/gemini-jailbreak-prompts): Detection and prevention for Google Gemini
- [Gemini 3 Jailbreak](https://wardstone.ai/jailbreaks/gemini-3-jailbreak): Detect and block attacks on Gemini 3 Pro, Flash, and Deep Think

### xAI
- [Grok Jailbreak](https://wardstone.ai/jailbreaks/grok-jailbreak): Detection and moderation for xAI Grok
- [Grok 4 Jailbreak](https://wardstone.ai/jailbreaks/grok-4-jailbreak): Detect and block attacks on Grok 4.1 and Grok 4

### Microsoft
- [Copilot Jailbreak](https://wardstone.ai/jailbreaks/copilot-jailbreak): Detection and prevention for Microsoft Copilot

### Meta
- [Llama Jailbreak](https://wardstone.ai/jailbreaks/llama-jailbreak): Detect and block attacks on Meta Llama
- [Llama 4 Jailbreak](https://wardstone.ai/jailbreaks/llama-4-jailbreak): Detect and block attacks on Llama 4 Scout and Maverick

### DeepSeek
- [DeepSeek Jailbreak](https://wardstone.ai/jailbreaks/deepseek-jailbreak): Detection and prevention for DeepSeek models
- [DeepSeek R1 Jailbreak](https://wardstone.ai/jailbreaks/deepseek-r1-jailbreak): Detection and prevention for DeepSeek R1 reasoning model

### General Guides
- [Prevent Prompt Injection](https://wardstone.ai/jailbreaks/prevent-prompt-injection): Complete guide to preventing prompt injection in LLM apps
- [Prompt Injection Defense](https://wardstone.ai/jailbreaks/prompt-injection-defense): Build robust prompt injection defenses

## LLM Provider Integrations

Step-by-step guides for securing AI applications with each provider:

### Major Providers
- [OpenAI](https://wardstone.ai/integrations/openai): Secure GPT-5.3-Codex, GPT-5.2, o3, and o4-mini applications
- [Anthropic Claude](https://wardstone.ai/integrations/anthropic): Protect Claude Opus 4.6, Opus 4.5, Sonnet, and Haiku
- [Google Gemini](https://wardstone.ai/integrations/google-gemini): Secure Gemini 3 Pro, Flash, and Deep Think models
- [Mistral AI](https://wardstone.ai/integrations/mistral): Protect Mistral Large 3, Devstral 2, and Magistral
- [Cohere](https://wardstone.ai/integrations/cohere): Secure Command A and RAG pipelines
- [DeepSeek](https://wardstone.ai/integrations/deepseek): Protect DeepSeek-V3 and R1 reasoning models
- [xAI Grok](https://wardstone.ai/integrations/xai-grok): Secure Grok 4.1 and Grok 4 with real-time data

### Cloud Platforms
- [Azure OpenAI](https://wardstone.ai/integrations/azure-openai): Enterprise-grade security for Azure deployments
- [AWS Bedrock](https://wardstone.ai/integrations/aws-bedrock): Unified security for multi-model deployments

### Open Source
- [Meta Llama](https://wardstone.ai/integrations/meta-llama): Protect Llama 4 Scout and Maverick models
- [Ollama](https://wardstone.ai/integrations/ollama): Secure local AI deployments

### Inference Platforms
- [Hugging Face](https://wardstone.ai/integrations/huggingface): Protect Inference Endpoints and Hub models
- [Replicate](https://wardstone.ai/integrations/replicate): Secure cloud model hosting
- [Together AI](https://wardstone.ai/integrations/together-ai): Enterprise open-source security
- [Groq](https://wardstone.ai/integrations/groq): Security at LPU speed (1200+ tok/s)
- [Perplexity](https://wardstone.ai/integrations/perplexity): Secure search-augmented AI

## Competitor Comparisons

Detailed comparisons of Wardstone vs other LLM security solutions:

- [Wardstone vs Lakera](https://wardstone.ai/compare/lakera): Compare features, accuracy, and pricing
- [Wardstone vs OpenAI Moderation](https://wardstone.ai/compare/openai-moderation): Free moderation API comparison
- [Wardstone vs Azure Content Safety](https://wardstone.ai/compare/azure-content-safety): Enterprise cloud comparison
- [Wardstone vs Llama Guard](https://wardstone.ai/compare/llama-guard): Open-source safety classifier comparison
- [Wardstone vs NeMo Guardrails](https://wardstone.ai/compare/nvidia-nemo): NVIDIA guardrails framework comparison
- [Wardstone vs Prompt Armor](https://wardstone.ai/compare/prompt-armor): LLM security startup comparison

## Industry Solutions

AI security tailored for specific industry compliance and use cases:

- [Healthcare](https://wardstone.ai/solutions/healthcare): AI security for healthcare applications and patient data protection
- [Fintech](https://wardstone.ai/solutions/fintech): AI security for financial services and fraud prevention
- [Legal](https://wardstone.ai/solutions/legal): Protect attorney-client privilege in legal AI tools
- [Education](https://wardstone.ai/solutions/education): AI security for educational applications and student data
- [Government](https://wardstone.ai/solutions/government): High-security AI for government applications
- [HR & Recruiting](https://wardstone.ai/solutions/hr-recruiting): Secure AI for hiring and recruitment
- [Customer Support](https://wardstone.ai/solutions/customer-support): Secure AI chatbots and agents
- [E-commerce](https://wardstone.ai/solutions/ecommerce): AI security for shopping and fraud prevention
- [SaaS](https://wardstone.ai/solutions/saas): Enterprise-ready AI features for B2B applications
- [Gaming](https://wardstone.ai/solutions/gaming): AI moderation for safe gaming environments

## API

- [Detection API](https://wardstone.ai/api/detect): POST endpoint accepting `{"text": "..."}` and returning `flagged`, `risk_bands`, `primary_category`, and threat metadata

## Integration

Wardstone provides drop-in protection via:
- REST API: `POST /api/detect` with JSON body
- TypeScript SDK: `await wardstone.guard(prompt)` ([npm](https://www.npmjs.com/package/wardstone))
- Python SDK: `wardstone.guard(prompt)` ([PyPI](https://pypi.org/project/wardstone/))
- Go SDK: `client.Guard(ctx, prompt)` ([GitHub](https://github.com/Wardstone-AI/wardstone-go))
- Ruby SDK: `client.guard(prompt)` ([RubyGems](https://rubygems.org/gems/wardstone))
- PHP SDK: `$client->guard($prompt)` ([Packagist](https://packagist.org/packages/wardstone/client))
- Java SDK: `client.guard(prompt)` ([Maven Central](https://central.sonatype.com/artifact/ai.wardstone/wardstone-client))
- C# SDK: `await client.GuardAsync(prompt)` ([NuGet](https://www.nuget.org/packages/Wardstone))

## Pricing

- **Free**: 10,000 API calls/month for testing and prototypes
- **Business**: $199/month for 500,000 calls with 99.9% uptime SLA
- **Enterprise**: Custom pricing for unlimited calls, on-premise deployment, and SSO