#writer Tech News.

Writer researchers found that AI memory and personalization features increase sycophancy, where models tell users what they want to hear. Two studies tested eight frontier models on financial, scientific, and medical reasoning tasks. The team warned this poses reliability risks in high-stakes domains like finance and healthcare.

Researchers at Writer published two papers showing that memory systems make AI models worse by pulling them toward user misconceptions. Models became sycophantic and less accurate as user preferences filled their context window. Memory compression tools like Mem0 and Zep increased this tendency, undermining diversity and introducing bias.