Detection Engineering Dispatch Prompted to Fail: When LLMs Go Rogue
Jun 18, 2025
37:30
LLMs are rewriting the rules of app security—and not always in a good way.
In this episode Alex sits down with Scott Rogers, a seasoned data scientist at ANvilogic to unpack why LLMs are the new wild west of application risk—and how old-school OWASP principles are making a serious comeback.
We cover:
- Real-world prompt injection failures (yes, including Air Canada’s rogue chatbot)
- How RAG systems can accidentally leak sensitive data
- Why GenAI risk ≠ traditional appsec—but it rhymes
- How classic tools like SAST, DAST, and logs can still save your bacon
Whether you're threat modeling your first LLM system or already knee-deep in GenAI, this episode is full of spicy detection ideas, war stories, and practical advice you won’t want to miss.
Stay in the loop! Connect with us:
- Join Dispatch Community: https://www.anvilogic.com/workshop
- Website: https://www.anvilogic.com/
- LinkedIn: https://www.linkedin.com/company/anvilogic
- YouTube: https://www.youtube.com/@Anvilogic
Detection Engineering Dispatch features candid conversations with security teams at top companies on how they build, measure, and scale world-class detection programs.