CogSLLaM: Cognitive Security for Large Language Models

Abstract

Large Language Models (LLMs) are increasingly employed across general purpose, expert, and domain-intensive user contexts; in all instances, users rely on these systems for information summarization, task planning, and problem-solving, and even for advice and recommendations related to decision-making. LLMs are becoming embedded in highly technical do-mains, supporting cyber security, code development, health care, organizational policy analysis, and the interpretation, specification, and implementation of complex technical requirements. This expanding range of applications underscores the growing role of LLMs as general-purpose cognitive and technical support tools. However, increasing reliance on LLMs also introduces cognitive security risks, as erroneous or mis-leading outputs can shape user understanding, judgment, and decision-making in ways that produce real-world consequences. In this paper, we examine cognitive security challenges in LLM applications, their downstream effects on users, relevant technical approaches for addressing these risks, representative use cases, and our proposed Cognitive Security for LLMs framework. This framework focuses on three dimensions along which cognitive security may be threatened: informational, se-mantic, and stylistic. We conclude with key takeaways and future directions for reducing misleading and potentially harmful chatbot interactions across every day and high-stakes contexts.