SEC536: Adversarial AI - Penetration Testing AI Systems


Generative AI is quickly becoming a default part of the software development lifecycle. Faster delivery can also increase security risk, operational fragility, and decisions that are difficult to review or audit after the fact.
This session focuses on AI coding guardrails: practical, developer-friendly controls that help teams adopt coding assistants with confidence while maintaining strong security, governance, and assurance. We will walk through four maturity levels for agentic coding controls covering topics like built-in steering mechanisms in modern agentic IDEs, project context files, RAG-powered context-injection, and how to implement deterministic, policy-as-code guardrails in that enforce security, quality, and compliance requirements with measurable outcomes and audit-ready evidence.
Then we will show how you can make guardrails that meet your company requirements, without annoying your developers. The secret is to allow them to evolve. Give agents a way to provide feedback on their guardrails and you'll quickly learn more about scenarios in your company that you never knew existed. And, you'll be able to accommodate them, automatically and with a human in the loop when you want it and full automation when you don't.
In-Person