Provides adaptive threat intelligence and runtime defense mechanisms to secure AI agents against various attacks.
Sponsored
Agent Immune acts as an adaptive immune system for AI agents, designed to complement existing governance toolkits rather than replace them. It learns from security incidents to detect and block rephrased or novel attacks that static rules might miss. The tool offers robust protection capabilities including semantic memory for catching indirect injections, multi-turn escalation detection, sensitive output scanning, rate limiting, and proactive prompt hardening, ensuring comprehensive security for AI interactions.
主な機能
01Built-in rate limiting and circuit breaker for session-level security
020 GitHub stars
03Multi-turn escalation detection for session-based threat tracking
04Adaptive semantic memory to detect rephrased and novel attacks
05Proactive prompt hardening with role-lock, sandboxing, and output guards
06Output scanning for PII, credentials, prompt leaks, and encoded blobs
ユースケース
01Securing AI chat APIs and applications against prompt injection and data exfiltration
02Integrating runtime defense into LangChain or CrewAI agents
03Protecting system prompts from malicious inputs and ensuring safe AI agent behavior
