SEC536: Adversarial AI - Penetration Testing AI Systems

In-Person
This keynote evaluates the cyber threat landscape across Jensen Huang’s five-layer AI cake—Energy, Chips, Infrastructure, Models, and Applications—and outlines strategic risk mitigations for enterprise leadership. While maintaining the focus on the EU region, the objective is to propose a general threat modeling framework that comprehends all the major dimensions of the AI adoption risk and proposes actionable indications on how to suitably manage them.
In-Person
AI has changed dramatically many areas. Cybersecurity is one of them. Searching, exploiting and patching vulns or managing cybersecurity is completely related to AI, but at the same time AI Digital Services and Agentic AI are bringing new set of vulnerabilities, therefore exploiting AI and securing AI is a new task that is changing dramatically the way CISOs manage the hardening of the systems. It's crazy moment for cybersecurity because of IA. Come to enjoy a talk about this brave new world. Come to this talk to enjoy with Jailbreak, Swarms, Prompt Injection, Fragmented Jailbreak Attacks, Misalignment, Poisoning, BIAS, Hallucinations, and what to to do with it if you work in Cyber.
In-Person
This presentation explores the rapidly evolving landscape of AI security, highlighting how widespread AI adoption has rendered older defenses obsolete and forced security teams to quickly adapt. Drawing on Cloudflare's observations, the talk distinguishes between the risk models of internal and public-facing applications, focusing primarily on securing public assets. It details Cloudflare's AI Security for Apps and notes a critical shift from mitigating basic prompt injections to managing custom topics, emphasizing that new defensive strategies are required as state-of-the-art LLMs outpace traditional guardrails. Furthermore, the session addresses the necessity of securing data access for AI agents utilizing emerging standards like the Model Context Protocol (MCP), briefly outlines the Cloudflare OS model for safe internal productivity and concludes with key actionable takeaways.
In-Person
In-Person