Mitigating Over-Personalization in LLMs via Structured Memory
Hakeem Hannoon, Andrew Zhao, Mihir Narayan, Sharvin Goyal, Ivaxi Sheth
cs.AI, cs.CL
Submitted: 2026-08-08
Updated: 2026-08-11
Code: https://github.com/toastedgrain/structured-memory
License: http://creativecommons.org/licenses/by/4.0/
The gist: Conversational assistants increasingly rely on persistent long-term memory to personalize responses across sessions.
Terminology
Abstract
Conversational assistants increasingly rely on persistent long-term memory to personalize responses across sessions. However, when stored user information is reintroduced into the model context, it can also influence responses in inappropriate or unrelated settings. We study two such failure modes in memory-augmented LLMs: cross-domain leakage, where memories from one life domain affect responses in another, and memory-induced sycophancy, where stored user beliefs make models more likely to agree with the user rather than respond truthfully. We apply a simple inference-time modification to how memories are presented to the model, without changing the model or the memory contents. Across seven models on PersistBench, we compare the commonly used all-in context format, where memories are injected as an unstructured list, with structured formats that partition memories by domain. This simple modification consistently reduces cross-domain leakage while preserving utility, with our strongest method reducing leakage by 8.8% on average relative to the baseline.
Sources
- DeepSeek-V3.2: Pushing the Frontier of Open Large Language Models
- Privacy Collapse: Benign Fine-Tuning Can Break Contextual Privacy in Language Models
- LLM Task Interference: An Initial Study on the Impact of Task-Switch in Conversational History
- Prompt Leakage effect and defense strategies for multi-turn LLM interactions
- AirGapAgent: Protecting Privacy-Conscious Conversational Agents
- Contextual Integrity in LLMs via Reasoning and Reinforcement Learning
- Hindsight is 20/20: Building Agent Memory that Retains, Recalls, and Reflects
- Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory
- REALTALK: A 21-Day Real-World Dataset for Long-Term Conversation
- Evaluating Very Long-Term Conversational Memory of LLM Agents
- Zep: A Temporal Knowledge Graph Architecture for Agent Memory
- Collaborative Memory: Multi-User Memory Sharing in LLM Agents with Dynamic Access Control
- CIMemories: A Compositional Benchmark for Contextual Integrity of Persistent Memory in LLMs
- RAPTOR: Recursive Abstractive Processing for Tree-Organized Retrieval
- gpt-oss-120b & gpt-oss-20b Model Card
- MemGPT: Towards LLMs as Operating Systems
- Hierarchical Memory for High-Efficiency Long-Term Reasoning in LLM Agents
- Generative Agents: Interactive Simulacra of Human Behavior
- Mitigating Memorization in LLMs using Activation Steering
- Kimi K2: Open Agentic Intelligence
Related papers
- MAVEN-T: Reinforced Heterogeneous Distillation for Real-Time Multi-Agent Trajectory Prediction
- Model Discovery Agent: LLM-assisted Bayesian experiment design for data-efficient discovery of mechanistic world models
- Agents' Last Exam
- The Clinician's Veto: Navigating Trust, Liability, and Uncertainty in Autonomous AI Prescribing
- MindHelper: Closed-Loop Embodied Mental-State Reasoning for Precision Intervention
- Incumbent Advantage: Brand Bias and Cognitive Manipulation Dynamics in LLM Recommendation Systems