AI Security Introduction

HipoCap Shield provides runtime protection for your AI applications, intercepting and blocking prompt injection attacks, sensitive keyword exposure, and unauthorized function access before they reach the LLM or the user.

Stop analyzing what went wrong. Prevent it from happening.

What is the Hipocap Shield?

The HipoCap Shield is the Enforcement Layer that sits between your AI application and function calls. Unlike traditional observability tools that analyze events after they occur, the Shield actively intercepts and blocks threats in real-time. It analyzes every function call request through three stages of security analysis before allowing execution, ensuring threats never reach your models or users.

Threat Categories

Hipocap detects threats across 14 categories (S1-S14):

• S1: Violent Crimes
• S2: Non-Violent Crimes
• S3: Sex-Related Crimes
• S4: Child Sexual Exploitation
• S5: Defamation
• S6: Specialized Advice (medical, legal, financial)
• S7: Privacy Violations
• S8: Intellectual Property Violations
• S9: Indiscriminate Weapons
• S10: Hate Speech
• S11: Suicide & Self-Harm
• S12: Sexual Content
• S13: Election Manipulation
• S14: Code Interpreter Abuse

Multi-Stage Analysis

Stage 1: Input Analysis

Uses Prompt Guard to detect malicious patterns in function inputs before execution.

Stage 2: LLM Analysis

Analyzes function results for threat indicators, attack patterns, and function call attempts.

Stage 3: Quarantine Analysis

Simulates infection by sending content to a quarantine LLM, then analyzes the output for hidden instructions.

Key Features

• Runtime Protection - Intercepts and blocks threats before they reach the LLM or user, not after
• Prompt Injection Protection - Detects and blocks indirect prompt injection attacks in real-time
• Keyword Detection - Identifies and protects sensitive keywords and patterns before they’re exposed
• Function Call Detection - Detects unauthorized function call attempts embedded in content and blocks execution
• Keyword Detection - Customizable keyword detection for sensitive patterns
• Real-time Analysis - Fast analysis with configurable quick mode for low-latency requirements—security checks don’t slow down your agents

How It Works

1. Function call is intercepted by Hipocap Shield
2. Multi-stage analysis runs (Input → LLM → Quarantine)
3. RBAC and governance rules are checked
4. Decision is made: ALLOW or BLOCK
5. All activity is traced and logged for observability

Next Steps

• Set up the Shield - Configure security analysis
• Keyword Detection - Configure keyword detection
• Prompt Injection Protection - Understand multi-stage analysis
• Threat Categories - Detailed reference for S1-S14 categories