mlsanityData to 14 Oct 2026

Fidelity Is Not Safety: Gently-Compressed LLMs Pass Every Data-Free Quality Guard Yet Invent Procedure Steps in Agentic Execution

arXiv 2607.28196 ↗First posted 30 Jul 2026cs.CLcited by 1

On the map: LLM agents and tool use → Agent memory systems

Abstract (arXiv)

Practitioners accept a compressed language model once it clears a stack of data-cheap quality guards: perplexity within a small factor of the original, downstream accuracy (for example MMLU) inside a confidence interval, and data-free output-fidelity signals that compare the compressed and original network's internal representations under random probe inputs. This stack has a blind spot. Across three model families, gently-compressed models clear every guard and then invent procedure steps that were never in the instructions when they run a standard operating procedure (SOP) as an agent. The effect is operator-specific: coherent low-rank (SVD) truncation induces it, and magnitude pruning matched to the same perplexity does not. One dissociation isolates the cause. The same compressed weights that CI-win a paired output-fidelity test CI-fail the invented-step canary. The governing axis is the coherence of the compression error times its rate; the magnitude of the damage does not predict it. The data-free fidelity probe is a fidelity oracle by construction, so it cannot see this axis. We characterize the blindspot and dissociation with paired confidence intervals on a pre-registered, powered canary across three architectures. Operator-specificity replicates on all three, and the perplexity-guard evasion appears where the model admits in-guard low-rank headroom. We then give a data-free screen: a two-axis statistic of the compression error (coherent-fraction and error-rate) that flags the failing builds with fixed thresholds across architectures and matches the coherence-times-rate mechanism. Perplexity, MMLU, and fidelity acceptance do not certify agent safety. Screen gently-compressed low-rank builds before agentic deployment

Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in Agent memory systems, one of 16 directions in LLM agents and tool use. Builds on 12 of its 13 references that are in the map. First posted on arXiv 30 Jul 2026.

Builds on

12 of 13 references.

References · 12

  1. 2022ReAct: Synergizing Reasoning and Acting in Language Modelscited by 8,695
  2. 2023Toolformer: Language Models Can Teach Themselves to Use Toolscited by 4,182
  3. 2022GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformerscited by 1,919
  4. 2019Similarity of Neural Network Representations Revisitedcited by 1,755
  5. 2022SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Modelscited by 1,386
  6. 2023AWQ: Activation-aware Weight Quantization for LLM Compression and Accelerationcited by 1,298
  7. 2023SparseGPT: Massive Language Models Can Be Accurately Pruned in One-Shotcited by 1,121
  8. 2023A Simple and Effective Pruning Approach for Large Language Modelscited by 800
  9. 2019HAWQ: Hessian AWare Quantization of Neural Networks with Mixed-Precisioncited by 402
  10. 2019HAWQ-V2: Hessian Aware trace-Weighted Quantization of Neural Networkscited by 261
  11. 2024SVD-LLM: Truncation-aware Singular Value Decomposition for Large Language Model Compressioncited by 243
  12. 2018Categorizing Variants of Goodhart's Lawcited by 149

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Quantization Error Is Spectrally Flat: A Single Random Probe Is a Calibrated, Data-Free Sensitivity Estimator, with Application to Budget-Targeted Mixed-Precision Quantizationcited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Does Compression Preserve Uncertainty? A Unified Benchmark for Quantized and Sparse LLMs via Conformal Prediction
  2. 2025Towards Understanding and Improving Refusal in Compressed Models via Mechanistic Interpretability
  3. 2024Beyond Perplexity: Multi-dimensional Safety Evaluation of LLM Compression
  4. 2025When Compression Becomes an Attack Surface: Black-Box Attacks on Prompt-Compressed LLM Agents
  5. 2024Decoding Compressed Trust: Scrutinizing the Trustworthiness of Efficient LLMs Under Compression
  6. 2026When Summaries Distort Decisions: Information Fidelity in LLM-Compressed Financial Analysis
  7. 2026Bayes-Sufficient Compression Is Not Enough: How Does Communication Help Multi-Agent Systems?
  8. 2026Reason Less, Verify More: Deterministic Gates Recover a Silent Policy-Violation Failure Mode in Tool-Using LLM Agents