mlsanityData to 13 Oct 2026

When Evidence Changes: Evaluating Memory Repair and Re-reading in Language-Model Agents

arXiv 2610.03902 ↗First posted 2 Oct 2026cs.CLcited by 0

On the map: LLM agents and tool use → Agent memory systems also fits: Medical and scientific domain agents

Abstract (arXiv)

When documents supporting an agent's derived facts are revoked or replaced, should it repair memory or re-read current evidence? We introduce an evidence-revision evaluation on medication- and problem-list tasks from public ICU records. Under revocation, replacement and control events, we compare full and source-filtered re-reading with caching, rebuilding and graph-local repair across two 7B models. Memory is supplied in full without retrieval, and costs include ingest, revision and every use. On short records, local repair uses 5-10× fewer revision tokens than rebuilding, yet every memory pipeline costs at least twice full re-reading in held-out conditions. In a small pre-specified development sweep, adding task-ineligible documents extended records to about 10,000 tokens; at that length, memory's mean cumulative cost fell below full re-reading's after 2-14 uses, partly through truncated extraction, while source-filtered re-reading remained cheapest. In the replacement study, none of the four primary confirmatory tests reached statistical significance. These results show why the cost of agent memory after evidence revision must be assessed against source-filtered re-reading over the full pipeline.

Key facts. No citing papers in the map yet (data through 13 Oct 2026). Placed in Agent memory systems, one of 16 directions in LLM agents and tool use. Builds on 33 of its 38 references that are in the map. First posted on arXiv 2 Oct 2026.

Builds on

33 of 38 references.

Closest ancestors · 16

Same direction, newest first

  1. 2026Revoked but Still Authoritative: An Empirical Study of Revocation Enforcement in Agent-Memory Systems8 Sep
  2. 2026Stored Is Not Supported: Typed Provenance and Assertion Guardrails for Persistent AI Agents2 Sep
  3. 2026Invalidation Contracts for Cross-Episode Agent Memory31 Aug
  4. 2026Total Recall at What Cost? Benchmarking the Serving Cost of Agentic Memory Systems12 Aug
  5. 2026From Faulty Memories to Corrected Actions: Dependency-Guided Rollback Repair for Memory-Augmented Agents11 Aug
  6. 2026When Memory Updates but Behavior Does Not: Repairing Implicit Stale Dependencies in Personalized Agent Responses3 Aug
  7. 2026MemTxn: A Transaction Boundary for Source-Supported Updates and Complete-State Recovery in Agent Memory30 Jul
  8. 2026MemTX: Transactional Belief Commit for Stateful Agent Memory27 Jul
  9. 2026When Does Memory Help? A Cost-Aware Evaluation of Long-Term Memory in Tool-Using LLM Agents26 Jul
  10. 2026Ground Truth First: A Longitudinal Evaluation Instrument for Agent Memory, and the Tenure Crossover in Memory-Architecture Rankings24 Jul
  11. 2026Supersede: Diagnosing and Training the Memory-Update Gap in LLM Agents25 Jun
  12. 2026Infini Memory: Maintainable Topic Documents for Long-Term LLM Agent Memory9 Jun
4 more
  1. 2026MEMOREPAIR: Barrier-First Cascade Repair in Agentic Memory8 May
  2. 2026STALE: Can LLM Agents Know When Their Memories Are No Longer Valid?7 May
  3. 2026Beyond the Context Window: A Cost-Performance Analysis of Fact-Based Memory vs. Long-Context LLMs for Persistent Agents5 Mar
  4. 2025Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory28 Apr

Other references · 17

  1. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,831
  2. 2024Qwen2.5 Technical Reportcited by 4,840
  3. 2023Mistral 7Bcited by 3,123
  4. 2024AgentPoison: Red-teaming LLM Agents via Poisoning Memory or Knowledge Basescited by 421
  5. 2025Defeating Prompt Injections by Designcited by 284
  6. 2025Securing AI Agents with Information-Flow Controlcited by 115
  7. 2026From Untrusted Input to Trusted Memory: A Systematic Study of Memory Poisoning Attacks in LLM Agentscited by 30
  8. 2026Cordon: Semantic Transactions for Tool-Using LLM Agentscited by 21
  9. 2026Securing LLM-Agent Long-Term Memory Against Poisoning: Non-Malleable, Origin-Bound Authority with Machine-Checked Guaranteescited by 17
  10. 2026MemLineage: Lineage-Guided Enforcement for LLM Agent Memorycited by 16
  11. 2026EvoArena: Tracking Memory Evolution for Robust LLM Agents in Dynamic Environmentscited by 9
  12. 2026Your Agents Are Aging Too: Agent Lifespan Engineering for Deployed Systemscited by 8
5 more
  1. 2026Verified Detection and Prevention of Concurrency Anomalies in Multi-Agent Large Language Model Systemscited by 8
  2. 2026When Agents Remember Too Much: Memory Poisoning Attacks on Large Language Model Agentscited by 6
  3. 2026TEPA: Revoking Stale Memories for Conflict-Robust Language Agentscited by 4
  4. 2026Agent Memory Is a Surface for Endogenous Authorization Launderingcited by 4
  5. 2026MemSecBench: Tracking Agent Memory Poisoning from Persistence to Consequence and Repaircited by 3

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Stale-Document Poisoning: When Outdated Retrieval Overrides Correct Model Answers
  2. 2026What Eviction Destroys: A Restore-Counterfactual Audit of Forgetting in Agent Memory
  3. 2026Looping Is Not Reliability: State-Bound Evidence and Typed Revision Contracts for Agentic Code Repair
  4. 2026PACMI: Provenance-Aware Cascading Memory Invalidation for Long-Term LLM Agents
  5. 2026Causal Episodic Memory for Feedback-Driven Agent Repair
  6. 2026When Stale Constraints Go Unchecked: Budgeted Verification Failures in Inherited Agent Memory
  7. 2026Memory Is a Derivation: The Distributed-Evidence Paradox in Long-Term Agents
  8. 2026Diagnosing Evidence Utilization in Long-Context and Retrieval-Augmented Language Models under Matched Evidence Conditions