Your AI systems are in production. Is your threat model? Prompt injection. Data poisoning. Model theft. These aren't theoretical risks on a researcher's blog anymore. They're active attack vectors hitting teams who thought their existing security frameworks had it covered. They didn't, because GenAI introduces a category of vulnerabilities that traditional tooling was never built to detect. This session covers how these attacks actually work in practice, where production AI systems are most exposed right now, and what defensive patterns are actually effective rather than just compliant. Real examples, real mitigations, and an honest look at where the current state of AI security is still falling short. If you're shipping AI in production and haven't stress-tested your threat model against these vectors, this is the session you cannot afford to skip.