ai-shield Runtime security for LLM and agentic AI systems. Detects prompt injection, covert channels, and C2-style behaviour in AI model traffic.