Large Language Models in Alzheimer’s Care: Clinical Use Cases, Safety Challenges, and Implementation Pathways

Abstract

Background: Alzheimer’s disease (AD) care is increasingly communication-intensive, requiring sustained caregiver education, symptom monitoring, behavioral management, and coordination across fragmented clinical settings. Large language models (LLMs) can generate, summarize, and adapt natural language at scale, creating opportunities to support dementia care workflows, but they also introduce safety risks that may be amplified in cognitively vulnerable populations. Methods: We provide a narrative synthesis of emerging applications of LLMs across the AD care continuum, examine how general LLM safety risks may be amplified or modified in Alzheimer’s care, and propose pragmatic implementation pathways for responsible clinical translation. Results: LLM use cases cluster into caregiver-facing support, patient-facing conversational agents (high caution), clinician workflow augmentation (highest near-term feasibility), clinical text intelligence for risk prediction (early-stage), and research/education support. Key safety threats include hallucinations, omission of critical information, over-reliance, bias, privacy leakage, and prompt-injection vulnerabilities. Conclusions: Implementation is most defensible using grounded architectures (e.g., retrieval-augmented generation), tiered deployment, human-in-the-loop verification, continuous monitoring, and security testing.