This paper explores the use of generative AI in combating misinformation on social media. The study presents three experiments: (1) an experiment with a simulated social media environment to measure the effectiveness of misinformation interventions generated by large language models (LLMs), (2) a second experiment with personalized explanations tailored to users' demographics and beliefs to reduce confirmation bias, and (3) an analysis of potential harms from personalized generative AI used for automated disinformation. The findings show that LLM-based interventions significantly improve user accuracy in labeling content as true or false, with improvements up to 47.6%. Users also prefer more personalized interventions when deciding on news reliability. The study also examines how personalization affects the effectiveness of explanations. Personalized explanations are found to be more helpful to users, with higher helpfulness scores compared to nonpersonalized ones. However, the study also highlights the potential danger of LLMs being used to generate personalized disinformation that is harder to detect, especially when targeting specific demographic groups. The paper discusses the broader implications of using LLMs for misinformation mitigation, emphasizing the need for accurate models and collaboration between policymakers, researchers, and engineers to ensure ethical use. The study contributes to the growing body of research on misinformation and highlights the potential of LLMs in creating scalable and effective interventions. However, it also raises concerns about the misuse of such technologies for generating targeted disinformation. The study's findings suggest that while LLMs can be powerful tools for combating misinformation, their effectiveness depends on the accuracy of the models and the ability to personalize interventions effectively.This paper explores the use of generative AI in combating misinformation on social media. The study presents three experiments: (1) an experiment with a simulated social media environment to measure the effectiveness of misinformation interventions generated by large language models (LLMs), (2) a second experiment with personalized explanations tailored to users' demographics and beliefs to reduce confirmation bias, and (3) an analysis of potential harms from personalized generative AI used for automated disinformation. The findings show that LLM-based interventions significantly improve user accuracy in labeling content as true or false, with improvements up to 47.6%. Users also prefer more personalized interventions when deciding on news reliability. The study also examines how personalization affects the effectiveness of explanations. Personalized explanations are found to be more helpful to users, with higher helpfulness scores compared to nonpersonalized ones. However, the study also highlights the potential danger of LLMs being used to generate personalized disinformation that is harder to detect, especially when targeting specific demographic groups. The paper discusses the broader implications of using LLMs for misinformation mitigation, emphasizing the need for accurate models and collaboration between policymakers, researchers, and engineers to ensure ethical use. The study contributes to the growing body of research on misinformation and highlights the potential of LLMs in creating scalable and effective interventions. However, it also raises concerns about the misuse of such technologies for generating targeted disinformation. The study's findings suggest that while LLMs can be powerful tools for combating misinformation, their effectiveness depends on the accuracy of the models and the ability to personalize interventions effectively.