The Attribution Blind Spot: Layerwise Trajectory Diagnostics for Source Reliance in Retrieval-Augmented Language Models
On the map: Retrieval-augmented generation → Agentic RAG, agents and memory also fits: RAG robustness and adaptive retrieval
Abstract (arXiv)
A retrieval-augmented model can match a document without relying on it. Controlled knowledge conflicts make source choice observable and let us ask a second question that prediction alone cannot answer: which internal-state properties define useful intervention directions? We study paired hidden-state changes with Latent Trajectory Shift (LTS), a signed projection onto a training-fitted first principal component (PC1), and keep verified training exposure separate from behavioral source choice. Across the evaluated conflicts, state-change magnitude is often the stronger predictor, whereas signed PC1 is the stronger selective controller: equal-norm interventions change source preference while better preserving non-target behavior, and the frozen direction transfers across the tested datasets and aligned model pairs. A same-system OLMo study further combines positive choice and control results with inconclusive exposure detection at the achieved power. The central result is a separation: representations that diagnose what a model will choose need not be the representations that best control that choice.
Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Agentic RAG, agents and memory, one of 15 directions in Retrieval-augmented generation. Builds on 5 of its 35 references that are in the map. First posted on arXiv 7 Oct 2026.
Builds on
5 of 35 references.
References · 5
- 2016Understanding intermediate layers using linear classifier probescited by 1,279
- 2023Representation Engineering: A Top-Down Approach to AI Transparencycited by 1,239
- 20242 OLMo 2 Furiouscited by 302
- 2022Attributed Question Answering: Evaluation and Modeling for Attributed Large Language Modelscited by 108
- 2024Quantifying reliance on external information over parametric knowledge during Retrieval Augmented Generation (RAG) using mechanistic analysiscited by 7
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2026The Attribution Blind Spot: Detecting When Language Models Rely on Memory Rather Than Retrieved Context
- 2026Detecting Is Not Resolving: The Monitoring Control Gap in Retrieval Augmented LLMs
- 2026Language Models Act on Hidden Valence
- 2026Context-Grounding Gains Are Mediated by Pre-existing Machinery: Auditing GRPO, SFT, and DPO
- 2026Grounding Without Corrective Control: Truth-Tracking Profiles for Large Language Models
- 2026TRACE: Trajectory Return Attribution and Contrastive Erasure for Multi-Turn Safety
- 2026Steering Safely or Off a Cliff? Rethinking Specificity and Robustness in Inference-Time Interventions
- 2026To Copy or Not to Copy: Copying Is Easier to Induce Than Recall