AI Guardrails:AI Guardrails product overview

Product offerings

AI Guardrails provides two offerings for distinct content safety scenarios:

• Guardrails protects AI applications against adversarial attacks and unsafe model outputs through configurable risk detection and multiple integration methods.

  • Threat Detection

    Detects regulatory violations (pornography, political sensitivity, violence, gore), PII and credentials, jailbreak attempts, injection attacks, file and link risks, and prompt extraction attempts. Verifies LLM outputs against ground truth sources. Supports invisible digital watermarks in AI-generated images and text for provenance tracking.

  • Customization

    Configure detection thresholds, risk levels, and filter words for industry compliance (finance, education). Train models on proprietary datasets for vertical use cases.

  • Integration Methods

    Supports RESTful API for custom applications, AI Gateway for centralized policy enforcement, WAF for edge-level prompt blocking, Model Studio for one-click Tongyi integration, and third-party platforms (Dify agents, OpenClaw plug-ins).

• Content Moderation provides multimodal content safety for platforms hosting user-generated content across social media, gaming, e-commerce, and media scenarios. Core features include the moderation API, and the console.

  • API-Based Moderation

    Detects spam, hate speech, violence, and ad violations in text. Scans images for adult content, violence, gore, and political sensitivity. Performs frame-by-frame video analysis with audio detection. Identifies inappropriate speech, violence, and political content in audio. Suitable for platforms with public content: video, live streaming, social media, e-commerce, forums, and CDN.

  • Management Console

    View moderation trends, category distributions, and latency metrics. Define custom violation rules and blocklists. Use pre-configured detection profiles for gaming, education, and social scenarios.

Use cases

Guardrails targets AI-specific threats in generative AI applications. Content Moderation handles UGC patterns in user-facing platforms.

• Guardrails (AI Application Protection) - Common use cases include:

  • AI chatbots — Prevent prompt injection that manipulates chatbots into disclosing policies or generating harmful advice.

  • AI content creation — Filter inappropriate AI-generated text, images, and videos before publication.

  • Code assistants — Scan AI-generated code for vulnerabilities, credential leaks, and malicious patterns.

  • Enterprise AI assistants — Detect manipulation attempts and sensitive data extraction across multi-turn conversations.

  • Model serving — Add safety guardrails to third-party models (OpenAI, Anthropic, open-source) without modifying weights.

• Content Moderation (UGC Safety) - Common use cases include:

  • Social networks — Real-time detection of pornographic, violent, and politically sensitive content in public feeds.

  • User profiles — Scan usernames, avatars, and bios at registration to block offensive or impersonating accounts.

  • Gaming — Monitor chat channels for toxicity, cheating ads, and bot spam. Detect harassment and underage grooming in direct messages.

  • Video platforms — Automated age-appropriate content rating. Real-time frame analysis and audio transcription for live broadcasts.

  • E-commerce — Detect counterfeit goods, prohibited items (weapons, drugs), and misleading descriptions. Validate product photos against quality guidelines, detect NSFW images in fashion listings. Monitor buyer-seller communications for fraud.

Pricing

AI Guardrails offers three billing methods: pay-as-you-go for variable workloads, resource plans for discounted pre-purchased capacity, and QPS expansion packages for higher rate limits.

• For information about Guardrails pricing, see Activation and billing overview.

• Content Moderation pricing: Activation and billing.

Guardrails vs. Content Moderation

Choose the service that matches your primary use case.

• When to Use Guardrails - Pure AI scenarios where content originates from or is processed by generative AI models:

  • Text-to-text generation (chatbots, writing assistants), text-to-image generation, code generation, AI agent actions (tool use, function calling), and training data sanitization.

  • Key differentiators: AI-specific threat detection (prompt injection, hallucinations, prompt crawlers), context-aware multi-turn analysis, and sub-second latency.

• When to Use Content Moderation - UGC scenarios where content is uploaded or created by end users:

  • Social media posts (text, images, videos), e-commerce product listings, gaming chat and user-created content, document sharing platforms, and live streaming.

  • Key differentiators: UGC-optimized detection (spam, fake reviews, impersonation), pre-configured scenarios (avatars, nicknames, public chat), and OSS integration for storage compliance.

• For hybrid scenarios, choose the service based on the primary use case:

  • Social app with AI-generated image uploads — primary activity is user-published content, so use Content Moderation.

  • AI companion chat (user inputs, AI responds) — this is an AI application, so use Guardrails.

Content Moderation 2.0 vs. 1.0

Content Moderation 2.0 adds pre-configured business scenarios and improves performance, label richness, configuration flexibility, and pricing over version 1.0.