Agentic AI governance beyond RAG

# Join us: Agentic AI Governance Beyond RAG

A technical deep-dive on zero-trust data access for multi-step AI agents on AWS
RAG solved data governance with a single trick: filter what goes into the vector index, and you're done. But agentic AI doesn't work that way. An agent might discover tables, read schemas, write its own SQL, query vector stores, and synthesize an answer — five steps, five opportunities for something to go wrong. One filter at retrieval can't govern a chain like that.
In this session, we'll walk through a production-grade architecture for governing agentic systems, built on AWS primitives:

• Why classic RAG governance breaks down the moment your system moves from "retrieve chunks" to "take actions"
• The Gateway pattern — how request/response interceptors act as a single choke point between agent and tools, enforcing scopes on every call and hiding unauthorized tools from the agent entirely
• Deterministic enforcement with Lake Formation — how row- and column-level security is enforced by the database engine itself, not by hoping the LLM behaves
• Defense in depth — five overlapping controls (workgroup limits, IAM, Lake Formation, Gateway interceptors, Bedrock Guardrails) that together mean no single point of failure
• The real trade-offs — what this buys you in security, and what it costs you in latency and token spend

Whether you're building agents that touch sensitive data, or just trying to figure out how "zero trust" applies once your AI system starts acting rather than just retrieving, this session will give you a concrete blueprint to work from.

Who should attend: engineers and architects building or securing agentic AI systems, especially on AWS; anyone evaluating how to move from RAG-based governance to something that holds up for multi-step agents.
We'll close with time for questions — bring your toughest governance problems.