MemTrapBench: Benchmarking Cognitive Traps in LLM Memory Use

Hugging Face Daily Papers Papers

Summary

This paper introduces MemTrapBench, a benchmark for evaluating cognitive traps in LLM memory use, and proposes AdaptiveMem, an inference-time method to mitigate these traps while preserving performance on standard benchmarks.

Memory has become a key component of large language models, enabling them to retain information and learn from long-term interactions. However, existing memory benchmarks mainly evaluate whether information is correctly extracted, stored, and retrieved, while largely overlooking how retrieved memories reshape model reasoning and affect performance on the current task. We identify memory-induced cognitive traps: even faithfully recorded and semantically relevant memories can distort model reasoning or beliefs and degrade current task performance. To systematically evaluate these failure modes, we introduce MemTrapBench, which covers two forms of cognitive traps: Reasoning Fixation and Belief Distortion. Experiments across two model families and five representative memory frameworks show that MemTrapBench is challenging: all evaluated memory strategies underperform the no-memory setting, with even the strongest methods suffering drops of more than 10%. To mitigate these cognitive traps, we propose AdaptiveMem, a simple yet effective inference-time method that instructs LLMs to avoid memory traps. AdaptiveMem mitigates cognitive traps on MemTrapBench while preserving or improving performance on standard memory benchmarks across diverse memory frameworks.
Original Article
View Cached Full Text

Cached at: 08/21/26, 04:08 AM

Paper page - MemTrapBench: Benchmarking Cognitive Traps in LLM Memory Use

Source: https://huggingface.co/papers/2608.20202

Abstract

Retrieved memories can induce reasoning errors and belief distortions in large language models, and an inference-time strategy helps avoid these cognitive traps while maintaining benchmark performance.

Memory has become a key component of large language models, enabling them to retain information and learn from long-term interactions. However, existing memory benchmarks mainly evaluate whether information is correctly extracted, stored, and retrieved, while largely overlooking how retrieved memories reshape model reasoning and affect performance on the current task. We identifymemory-induced cognitive traps: even faithfully recorded and semantically relevant memories can distort model reasoning or beliefs and degrade current task performance. To systematically evaluate these failure modes, we introduceMemTrapBench, which covers two forms of cognitive traps:Reasoning FixationandBelief Distortion. Experiments across two model families and five representativememory frameworksshow thatMemTrapBenchis challenging: all evaluated memory strategies underperform the no-memory setting, with even the strongest methods suffering drops of more than 10%. To mitigate these cognitive traps, we proposeAdaptiveMem, a simple yet effectiveinference-time methodthat instructs LLMs to avoid memory traps.AdaptiveMemmitigates cognitive traps onMemTrapBenchwhile preserving or improving performance on standard memory benchmarks across diversememory frameworks.

View arXiv pageView PDFGitHub1Add to collection

Get this paper in your agent:

hf papers read 2608\.20202

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2608.20202 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2608.20202 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2608.20202 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

MemTrace: Probing What Final Accuracy Misses in Long-Term Memory

arXiv cs.AI

MemTrace is a benchmark that evaluates LLM agent memory at the knowledge point level, probing how facts behave under varying memory age, question type, and evidence conditions. It reveals that pooled accuracy hides distinct failure modes, and that the main bottleneck is evidence use rather than retrieval.

MemEvoBench: Benchmarking Memory MisEvolution in LLM Agents

arXiv cs.CL

MemEvoBench introduces the first benchmark for evaluating memory safety in LLM agents, measuring behavioral degradation from adversarial memory injection, noisy outputs, and biased feedback across QA and workflow tasks. The work reveals that memory evolution significantly contributes to safety failures and that static defenses are insufficient.