Enkrypt AI Guardrails is a runtime security layer that provides real-time control and enforcement for AI agents, tools, RAG systems, and Model Context Protocol (MCP) implementations. The product operates at multiple enforcement boundaries including prompt inputs, retrieval operations, tool execution, and outputs. The platform addresses security risks such as prompt injection (direct and indirect), tool misuse, privilege escalation, and data exfiltration. It also covers brand safety concerns including jailbreaks, toxicity, and deceptive behavior, as well as compliance requirements related to PII/PHI/PCI handling and data minimization. Guardrails enforces policies across five key boundaries: prompt filtering and injection defense, RAG source filtering with identity-aware retrieval, tool call approval/denial with allowlist/denylist capabilities, output tone and safety enforcement, and multimodal policy consistency across text, image, and audio inputs. The system logs all enforcement decisions with policy context to support incident response, audit reporting, and continuous improvement. It integrates with existing LLM platforms via API, supports agent orchestrators through hooks and middleware, and can enforce policies inline via MCP Gateway. Identity integration is available through SSO/IAM systems including Okta and Azure AD. The product is designed for low-latency operation with sub-15ms response times and includes multimodal and multilingual support. It complements red teaming activities by preventing discovered failure modes at runtime.