STALE: Can LLM Agents Know When Their Memories Are No Longer Valid?

TL;DR

This paper introduces STALE, a benchmark for evaluating LLMs' ability to detect and adapt to invalidated memories caused by implicit conflicts, highlighting a significant gap in current models' capabilities.

Contribution

The paper presents STALE, a comprehensive benchmark with 400 conflict scenarios, and evaluates LLMs' performance, revealing substantial challenges in memory revision and state awareness.

Findings

LLMs achieve only 55.2% accuracy in detecting outdated memories.

Models often accept outdated assumptions in user queries.

Explicit state adjudication improves memory revision capabilities.

Abstract

Large Language Model (LLM) agents are increasingly expected to maintain coherent, long-term personalized memory, yet current benchmarks primarily measure static fact retrieval, overlooking the ability to revise stored beliefs when new evidence emerges. We identify a critical and underexplored failure mode, Implicit Conflict: a later observation invalidates an earlier memory without explicit negation, requiring contextual inference and commonsense reasoning to detect. To rigorously evaluate this capability, we introduce STALE, a benchmark of 400 expert-validated conflict scenarios (1,200 evaluation queries across three probing dimensions) spanning over 100 everyday topics with contexts up to 150K tokens. We propose a three-dimensional probing framework that tests State Resolution (detecting that a prior belief is outdated), Premise Resistance (rejecting queries that falsely presuppose a stale state), and Implicit Policy Adaptation (proactively applying updated states in downstream behavior). A systematic evaluation of frontier LLMs and specialized memory frameworks reveals a pervasive gap between retrieving updated evidence and acting on it, with even the best evaluated model achieving only 55.2% overall accuracy. Models often accept outdated assumptions embedded in a user's query, and they struggle to recognize when a change in one aspect of the user's state should invalidate related memories. To establish an initial baseline for state-aware memory, we further present CUPMem, a prototype that strengthens write-time revision through structured state consolidation and propagation-aware search, suggesting that explicit state adjudication is a promising direction for robust agentic memory.