Tiny, Smart, and Infected: The Security Risks of the Agentic Future - 28 de enero de 2026 - TecnoWebinars.com

As the AI industry pivots from massive, generalist behemoths to efficient, specialized Small Language Models (SLMs) powering agentic workflows, we are inadvertently expanding our attack surface. This talk dissects the hidden security crisis within the modern AI supply chain, specifically how essential scaling techniques like LoRA adapters and fine-tuning serve as vectors for poisoned training data, "Trojan" model weights, and arbitrary code execution hidden within tensor file headers. By examining the lifecycle of a compromised agent—from the ingestion of malicious datasets to the deployment of infected LoRAs—we will uncover why the future of agentic AI requires a zero-trust approach to the very artifacts that make it intelligent.