The Generative AI Security Podcast Exploring GenAI Security: Agentic Top 10, Threat Modeling & Community Hackathons
12 snips
Jul 31, 2025 John Sotiropoulos, co-lead of the Agentic Security Initiative, dives into the cutting-edge realm of AI agent security. He discusses the powerful Agentic Top 10, detailing key vulnerabilities like memory and data poisoning. Listeners learn about the initiative's 15-threat taxonomy and how community hackathons are shaping real-world security insights. Sotiropoulos highlights opportunities for audience participation, including contributions on GitHub and upcoming events at Black Hat, making it clear that everyone can play a role in the future of safe AI development.
AI Snips
Chapters
Transcript
Episode notes
Practical Taxonomy For Agentic Security
- The Agentic Security Initiative created a practical taxonomy and 15 threats to make agentic AI security tangible.
- The team focused on usable threat models, mitigations, and graphical navigators for practitioners.
Use Hackathons To Validate Guidance
- Engage the community through hands-on events like hackathons and CTFs to collect real feedback.
- Use those events to convert theory into practical improvements and iterate on guidance.
Blend Methodology With Community Data
- The initiative combined threat modeling methodologies like MAESTRO with community-sourced data.
- This hybrid approach produced widely adopted documents and practical playbooks.