Context Contamination at USENIX Security 26 — Oral Presentation & Poster
Abstract
I presented our latest research at USENIX Security 2026 on August 12th. Our paper, “Context Contamination in LLM Analysis of Network Security Logs,” reveals how adversaries can embed malicious payloads into log data to trick the LLMs used by Security Operations Centers (SOCs), and introduces LogInject, a systematic framework to evaluate and mitigate these vulnerabilities. In addition to the oral presentation, we also presented a poster. This work is a cross-collaboration between PayPal AI Labs, University of Houston, and Kent State University.
