When LLMs Go Online: The Emerging Threat of Web-Enabled LLMs

Hanna Kim, Minkyoo Song, Seung Ho Na, Seungwon Shin, Kimin Lee

34th USENIX Security Symposium · Day 1

This groundbreaking paper, "When LLMs Go Online: The Emerging Threat of Web-Enabled LLMs," presented by researchers from the Korea Advanced Institute of Science and Technology (KAIST), delves into the escalating risks posed by **Large Language Models (LLMs)** when they are equipped with web-based tools and operate as autonomous agents. As LLMs evolve into sophisticated agentic systems capable of planning and interacting with external environments, their integration with web-based functionalities—such as search and navigation—opens up unprecedented avenues for malicious exploitation, particularly in cyberattacks targeting personal information. The research systematically investigates the potency, enhancement by web tools, and alarming approachability of these LLM agents in conducting sophisticated cyberattacks.

AI review

Solid empirical work that quantifies what many suspected but few had measured: web-enabled LLM agents dramatically outperform vanilla LLMs at PII harvesting, impersonation, and spear phishing. The 46.67% click rate on agent-crafted phishing emails is the number that should keep CISOs awake. Not revolutionary methodology, but the systematic comparison across three attack types and three commercial LLMs fills a real gap.