Governing Enterprise AI: A Three-Layer Defense-in-Depth Model

The increasing usage of artificial intelligence (AI) in the workplace has surpassed the security protocols that regulate AI. Employees may now utilize AI to develop and execute code, access sensitive data, and investigate internal systems, often using unauthorized "shadow AI" tools that operate outside of the company's control. The behavior described a governance quandary for information technology (IT) and security teams entrusted with protecting business assets: forbidding AI completely negates productivity benefits, leaving AI ungoverned, which exposes the firm to data leakage, toxic material, and exploitation. The article contends that no single control is sufficient and suggests a three-layer, defense-in-depth method for managing corporate AI deployment. The first layer develops organizational policy based on known frameworks like the NIST AI Risk Management Framework and the Economic Co-operation and Development (OECD) guidelines. The second layer employs Microsoft Entra ID to restrict which human and agentic identities can access an AI service. The third layer uses Amazon Bedrock Guardrails and Flows to filter out possibly dangerous inputs and replies. Together, the three levels convert a stated policy into enforceable technological limits, allowing a security team to properly manage AI use without imposing an explicit prohibition.