HitchHacker's Guide to Building Secure Agents - Kenneth Yeung & Kasimir Schulz

This talk was recorded at NDC AI in Oslo, Norway. #ndcai #ndcconferences #developer #softwaredeveloper Attend the next NDC conference near you: https://ndcconferences.com https://ndc-ai.com/ Subscribe to our YouTube channel and learn every day: / ‪@NDC‬ Follow our Social Media!   / ndcconferences     / ndc_conferences     / ndc_conferences   #ai #aiagents Multiple layers of guardrails. Humans in the loop. Zero-trust intent validation. We’ve all heard of these being essential to implement when designing an agentic system, but they add a lot of latency, worsen user experience, and are often too complex to efficiently maintain. In a world full of frameworks, policies, and AI security suites, how does one decide which are worth implementing? This talk will cover designing secure agents, how the various failure points are exposed in the real world, and how these failure points can be remediated. We’ll look into common misconceptions about security systems, and how key elements of secure agent design that are often overlooked. We’ll use the perspective of an AI attacker to understand where agentic systems are vulnerable, how to minimze the impact of these attacks, and how to make sure these impacts don’t make your agent utterly useless to its users.