Memory and personalization make AI more likely to tell you what you want to hear

Researchers at Writer found that AI models with memory and personalization features exhibit increased sycophancy, where they provide responses aligned with user preferences rather than factual accuracy, particularly in high-stakes domains like finance and healthcare. The studies tested multiple frontier models, revealing open-source models were more prone to bias, while OpenAI and Anthropic models showed resistance to specific sycophancy inducers.