Tutorial: Guarding the Gates: Understanding and Mitigating OWA...
summary
OWASP lists the Top 10 LLM risks and ranks prompt injection as the most critical, demonstrated by live Gandalf capture-the-flag lab and Viz/MLLDoc exercises where participants only reach flags three or four before guardrails break; real incidents include a chatbot selling a car for $1 and an airline honoring a bogus ticket. Mitigations shown include NeMo Guardrails, content segregation, human-in-the-loop, policy-as-code, plus privacy tools like differential privacy, confidential enclaves and homomorphic encryption.
AI-written summary · about
related talks
28mSecuring LLMs: Insights into OWASP top 10
20mSecurity Threats in Modern LLM Applications
54mA Practical Guide to AI Threats & Mitigation Strategies
59mSecuring the Digital Landscape: A Deep Dive into OWASP Top 10 for Applications, APIs, and LLMs
30mNavigating the New AI Security Landscape
20mUnderstanding and Mitigating Risks in LLM-Powered Web Apps
27mSecuring The Gateway And Mitigating Risks In LLM API Integration
26mOne guardrail is not enough: Defense-in-depth against LLM jailbreaks