mlsanityData to 7 Oct 2026

Looking Is Not Picking: An Attention-Segment Account of Tool-Selection Failures in LLM Agents

arXiv 2606.16364 ↗First posted 15 Jun 2026cs.AI cs.CR cs.SEcited by 3

On the map: LLM agents and tool use → Tool learning and function calling

Abstract (arXiv)

LLM agents mis-call tools, and the natural guess is that the model failed to see the right tool in a crowded harness. We show the opposite through a lens concurrent work sets aside -- the model's attention to labeled tool-definition segments. On real BFCL failures, by per-candidate attention argmax the model attends most to the correct tool 80% of the time (vs. 21% chance), and the gold is the under-attended segment on only 10%: it looks at the right tool and still picks wrong. This directly refutes the intuitive "crowded-harness / lost-in-the-middle" explanation: the failure is at the decision readout, not the harness, and we pin it there three ways. (1) Input vs. readout: repairing the prompt (reordering or duplicating the gold tool) recovers <=23% of failures, while readout-side interventions recover 59-91%. (2) Representation-invariance: two gold-pointed interventions in different representations -- an additive attention-logit bias and a residual-stream steering vector -- recover largely the same failures (per-task Jaccard 0.865 pooled, 0.79-0.91 per model), so the bottleneck is localized to the readout independent of which representation is poked. (3) A training-free, gold-free selector: per-segment attention closes most of the gold-free-vs-oracle gap on BFCL (+11.9 pts pooled function-name selection vs. +17.9-pt oracle headroom) and adds +14.9 pts on Seal-Tools; every model positive (exact McNemar p<=8e-4 each). Scopes differ: the causal attention-bias dose-response is bidirectional and monotonic on 10 mask-honoring models (3-32B), the full 0.5-32B span carrying only the correlational diagnostic; the deployable selector is evaluated on 5 single-turn models and does not yet transfer to a multi-turn loop.

Key facts. Cited by 3 papers in the map (data through 7 Oct 2026). Placed in Tool learning and function calling, one of 16 directions in LLM agents and tool use. Builds on 27 of its 39 references that are in the map. First posted on arXiv 15 Jun 2026.

Builds on

27 of 39 references.

Closest ancestors · 4

Same direction, newest first

  1. 2026Model-Adaptive Tool Necessity Reveals the Knowing-Doing Gap in LLM Tool Use13 May
  2. 2026Tool Calling is Linearly Readable and Steerable in Language Models8 May
  3. 2024Seal-Tools: Self-Instruct Tool Learning Dataset for Agent Tuning and Detailed Benchmark14 May
  4. 2023ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIs31 Jul

Other references · 23

  1. 2017Attention Is All You Needcited by 59,932
  2. 2024The Llama 3 Herd of Modelscited by 14,192
  3. 2022ReAct: Synergizing Reasoning and Acting in Language Modelscited by 8,355
  4. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,767
  5. 2023Lost in the Middle: How Language Models Use Long Contextscited by 3,383
  6. 2023Efficient Streaming Language Models with Attention Sinkscited by 2,115
  7. 2024Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phonecited by 1,883
  8. 2024τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domainscited by 1,280
  9. 2019Attention is not Explanationcited by 916
  10. 2020Quantifying Attention Flow in Transformerscited by 759
  11. 2024Yi: Open Foundation Models by 01.AIcited by 706
  12. 2019Attention is not not Explanationcited by 629
11 more
  1. 2024AttnLRP: Attention-Aware Layer-Wise Relevance Propagation for Transformerscited by 116
  2. 2025Which Attention Heads Matter for In-Context Learning?cited by 61
  3. 2025Detecting High-Stakes Interactions with Activation Probescited by 45
  4. 2024Attention in Large Language Models Yields Efficient Zero-Shot Re-Rankerscited by 42
  5. 2023Fortify the Shortest Stave in Attention: Enhancing Context Awareness of Large Language Models for Effective Tool Usecited by 27
  6. 2025Spotlight Your Instructions: Instruction-following with Dynamic Attention Steeringcited by 18
  7. 2024Mixture of In-Context Experts Enhance LLMs' Long Context Awarenesscited by 17
  8. 2026Accurate Failure Prediction in Agents Does Not Imply Effective Failure Preventioncited by 11
  9. 2025MindGuard: Intrinsic Decision Inspection for Securing LLM Agents Against Metadata Poisoningcited by 10
  10. 2026Beyond the Black Box: Interpretability of Agentic AI Tool Usecited by 3
  11. 2026Spectral Guardrails for Agents in the Wild: Detecting Tool Use Hallucinations via Attention Topologycited by 2

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 3.

3 papers.

  1. 2026SkillGate: Training In-Policy Skill Selection in Long-Horizon Agentscited by 1
  2. 2026When Retrieval Metrics Mislead: Measuring Policy Signal in Long-Horizon Tool-Use Agentscited by 0
  3. 2026The Default Trap: Rethinking Plan Evaluation in Tool-Using LLM Agentscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026LLM Agents Already Know When to Call Tools -- Even Without Reasoning
  2. 2026Memory-Induced Tool-Drift in LLM Agents
  3. 2026Loud Failures, Quiet Failures: Fault Detection and Recovery in Tool-Using Language Model Agents
  4. 2026Diagnosing Tool-Selection Reasoning in LLM Agents with Canary Tools
  5. 2026Set-shifting Behavioral Test for Harnessed Agents
  6. 2026Tunable Tool-Call Rates in LLM Agents via Representation Steering
  7. 2026ToolMenuBench: Benchmarking Tool-Menu Filtering Strategies for Reliable and Efficient LLM Agents
  8. 2026A Few Neurons Reveal When LLMs Misuse Tools: Sparse Detection and Selective Steering for Reliable Tool Use