BSidesSF 2026 - Security for AI Agents Using an Ensemble of Fine-tuned... (Lidan Hazout, Bar Kaduri)

This talk addresses the growing security risks associated with AI agents as they gain access to sensitive data and tools. The speakers, Lidan Hazout and Bar Kaduri, introduce a runtime guardrail that utilizes fine-tuned small language models (SLMs) to secure these interactions before tool invocation. They present benchmarks on leading open-source models and conduct live demonstrations of attack detection, highlighting the importance of robust security measures in AI applications.