SECURING THE AI-FIRST SDLC: IDENTITY, ACCESS, AND DATA PROTECTION FOR ENTERPRISE AI AGENTS

AI agents can now participate in several steps of software delivery: reading requirements, changing code, requesting tests, and preparing releases. Giving them this freedom creates a difficult security question. How can an enterprise allow useful work without giving an agent more authority than the task requires? Existing practices such as least privilege, identity management, separation of duties, and audit logging provide much of the foundation. The challenge is applying them consistently when a human delegates work to agents that use several connected systems. This paper proposes Agentic Identity and Access Governance (AIAG), a security framework for the AI-First Software Development Lifecycle (SDLC). Four connected mechanisms define the proposal: a Role Conflict Matrix that checks the combined permissions of people and their sponsored agents; an Agent Access Manifest that records permitted tools and operating limits; a governed boundary between live personal-information access and general AI-assisted work; and an MCP Gateway that enforces access policy before tool calls execute. A customer notification-preference change is used to explain how these mechanisms affect development, testing, release approval, and support. We also describe credential handling, data-egress controls, audit, and practical ways to evaluate the framework. The work is conceptual, informed by technical analysis and exploratory proof-of-concept activity. No end-to-end deployment or measured security results are claimed.