AI is shipping to production faster than most security programs can test it, and attackers know it. Valid AI vulnerability reports jumped 210% year over year on HackerOne's platform. Prompt injection alone surged 540%. Meanwhile, ~97% of AI-related incidents still trace back to basic access-control flaws.
The gap between how fast teams build AI and how fast they can prove it's safe is widening. This session shows how a company at the heart of the GenAI stack is closing it.
In 45 minutes, you'll walk away with:
- A real-world blueprint for securing AI-adjacent infrastructure on AWS — from a company thousands of AI apps are built on
- Which AI risks actually show up in production (prompt injection, retrieval-layer data exfiltration, agentic tool misuse) and which matter most, backed by HackerOne platform data
- How to combine continuous Bug Bounty, Pentest, code security, and agentic remediation into one exposure-management loop
- How to validate AI systems adversarially — mapped to the OWASP Top 10 for LLMs, MITRE ATLAS, and NIST AI RMF
- How to preserve developer velocity while adding AI-specific security testing (without becoming the "team that says no")
- One thing you can do Monday morning to close your AI security gap
