IAM has a reasoning problem. For decades, the industry has operated on hard-coded policies, manual system mapping, and access reviews that exist mainly to satisfy auditors. The arrival of AI has generated enormous hype, but bolting a general-purpose LLM onto legacy architecture doesn't change the underlying problem. It just produces faster noise.

This session draws on lessons from building AI-native IAM from the ground up at Opti. Specifically, why general-purpose models are insufficient for identity and what it actually takes to build specialized AI that understands entitlements, risk context, and lifecycle decisions with precision. We will walk through four purpose-built identity models: entitlements comprehension, risk mitigation, lifecycle management, and integrations, and what each revealed about the gap between AI that augments existing workflows and AI that fundamentally replaces broken ones.

The session addresses real architectural tradeoffs: normalizing identity language across fragmented systems, replacing confidence-free rubber-stamping with dynamic approvals, and building integrations that scale without custom services work.  Attendees will leave with a sharper lens for evaluating AI claims across the IAM vendor landscape and a clearer picture of what AI-native actually requires.