Confidently Wrong, Silently So: Auditing Undetectable Failures of a Deployed On-Device Language Model

• URL: http://arxiv.org/abs/2608.23663v2
• Date: Wed, 26 Aug 2026 17:45:35 GMT
• Title: Confidently Wrong, Silently So: Auditing Undetectable Failures of a Deployed On-Device Language Model
• Abstract summary: We present a reproducible reliability audit of the developer-accessible on-device foundation model.<n>We find a emphtask-asymmetric miscalibration: its guardrails fail in opposite directions across tasks.<n>We contribute a model-agnostic audit protocol, a surface-indistinguishability test, and released code and frozen evaluation items as reusable infrastructure for auditing deployed models.
• Score: 0.038379177968040606
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Aligning deployed language models requires knowing when their outputs can be trusted, yet on-device models now ship to hundreds of millions of devices with no server-side moderation, and the configuration developers can actually deploy is rarely audited independently. We present a reproducible reliability audit of the developer-accessible on-device foundation model, framed as an oversight question: can a user or a resource-constrained developer tell when the model is wrong? Red-teaming it on calibration, confident confabulation on false-premise questions, and over-refusal of benign prompts, we find a \emph{task-asymmetric miscalibration}: its guardrails fail in opposite directions across tasks (confabulating on 69\% of false premises while refusing 18\% of entirely benign inputs), atop a self-reported confidence that is saturated and non-discriminative (AUROC 0.47; ECE 70, worst among comparable small models). Crucially, confident-correct and confident-wrong outputs are \emph{surface-indistinguishable}: a classifier over 15 user-visible features separates them at AUROC only 0.55 (equivalence-confirmed), leaving no signal for oversight at inference time. No cheap single-generation signal flags these failures ($\le$0.68 AUROC), whereas a black-box consistency wrapper requiring no model access recovers reliability (confident confabulation 75\%$\to$3\%; selective accuracy 43\%$\to$83\%) at a tunable cost. We contribute a model-agnostic audit protocol, a surface-indistinguishability test, and released code and frozen evaluation items as reusable infrastructure for auditing deployed models.

Related papers

• Recognition Without Enforcement: Configuration-Dependent Failures in LLM Agent Instruction Arbitration and External Control [0.0]
  LLM agents arbitrate among instructions from system prompts, users, memory, and tools, but this arbitration cannot be assumed to enforce trust boundaries.<n>We identify a recognition-enforcement gap: source-format features are linearly decodable from model activations, and models can explicitly identify forged authority when prompted.<n>We treat model self-arbitration as a capability rather than a security boundary and implement an external reference monitor combining authenticated source routing with capability-gated tool execution.
  arXiv  Detail & Related papers  (2026-08-28T16:34:05Z)
• Open-Weight Masked Introspection: Measuring What Language Models Can Report About Their Own Computation [8.970269049715933]
  Recent work suggests that a complex enough model can audit its own internals, call out what changed, and report back confidently about it.<n>We tested that claim on eight open-weight models from seven families and found no such ability.<n>To test it we built Open-Weight Masked Introspection (OWMI), a framework that intervenes on residual-stream sites, attention heads and sparse-autoencoder features.
  arXiv  Detail & Related papers  (2026-08-20T21:09:50Z)
• Cybersecurity Detection Classification with Reasoning-enabled Language Models [50.72675435043546]
  A major issue in Security Operations Centers (SOCs) is alert fatigue.<n>Prior work prompts or fine-tunes large language models (LLMs) to emit a triage label directly, but does not train them to reason about whether a detection is a genuine threat.<n>We train a chain-of-thought (CoT) reasoning-enabled triage classifier on real, human-labeled Windows endpoint detections.
  arXiv  Detail & Related papers  (2026-07-30T16:22:13Z)
• Self-Authored Verification Is Unreliable in Heuristic Self-Improving Agents [5.91817208489443]
  We study the verifier-acceptance-deployment gap.<n>This gap refers to the discrepancy between an agent's self-authored verification signal and a sealed deployment evaluation.<n>We introduce a Sealed Exogenous Acceptance Loop (SEAL)
  arXiv  Detail & Related papers  (2026-07-27T11:45:14Z)
• Calibrated e-CUSUM Decoding for Quantized Reasoning Models: Why Token Log-Probability Is the Wrong Observable for Decoding Monitors [0.6999740786886536]
  Low-bit quantization makes small reasoning models inexpensive to deploy but can degrade their chains of thought.<n>We show that the centered token log-probability increment $log p(w_t)+H_t$ is the wrong observable for this purpose.<n>We introduce a training-free decoding controller that combines a degeneration-aware alarm score fusing token uncertainty with explicit repetition.
  arXiv  Detail & Related papers  (2026-07-13T09:34:36Z)
• CIAware-Bench: Benchmarking Control Intervention Awareness Across Frontier LLMs [100.38986535324284]
  We introduce textbfCIAware-Bench, a benchmark for measuring textbfcontrol textbfintervention (CI) awareness across frontier models.<n>CIAware-Bench tests whether models can distinguish their own trajectories from those modified by a control intervention.
  arXiv  Detail & Related papers  (2026-06-09T16:24:16Z)
• What Single-Prompt Accuracy Misses: A Multi-Variant Reliability Audit of Language Models [0.0]
  Single-prompt accuracy is the dominant way to benchmark language models, but it can miss reliability failures that matter.<n>We evaluate a 15-model open-weight corpus, with the main reliability analyses focused on 10 instruct models across five classification and reasoning benchmarks.<n>We find three broad results. First, evaluation design can materially change the conclusion.<n>Second, confidence signals are fragile. On MMLU-Pro, every primary model verbally reports confidence substantially above both its accuracy and its token-probability confidence on the same rows, and verbal parse rate can collapse for a single model on a single prompt variant.
  arXiv  Detail & Related papers  (2026-05-03T20:05:08Z)
• CoVerRL: Breaking the Consensus Trap in Label-Free Reasoning via Generator-Verifier Co-Evolution [52.691495954442985]
  CoVerRL is a framework where a single model alternates between generator and verifier roles, with each capability bootstrapping the other.<n> Experiments across Qwen and Llama model families demonstrate that CoVerRL outperforms label-free baselines by 4.7-5.9% on mathematical reasoning benchmarks.<n>Self-verification accuracy improves from around 55% to over 85%, confirming that both capabilities genuinely co-evolve.
  arXiv  Detail & Related papers  (2026-03-18T14:38:55Z)
• A Coin Flip for Safety: LLM Judges Fail to Reliably Measure Adversarial Robustness [57.510025257780306]
  We show that existing validation protocols fail to account for substantial distribution shifts inherent to red-teaming.<n>We propose ReliableBench, a benchmark of behaviors that remain more consistently judgeable, and JudgeStressTest, a dataset designed to expose judge failures.
  arXiv  Detail & Related papers  (2026-02-04T15:13:35Z)
• Trusted Uncertainty in Large Language Models: A Unified Framework for Confidence Calibration and Risk-Controlled Refusal [31.458406135473805]
  We present UniCR, a unified framework that turns heterogeneous uncertainty evidence into a calibrated probability of correctness.<n>UniCR learns a lightweight calibration head with temperature scaling and proper scoring.<n>Experiments on short-form QA, code generation with execution tests, and retrieval-augmented long-form QA show consistent improvements in calibration metrics.
  arXiv  Detail & Related papers  (2025-09-01T13:14:58Z)
• Are You Getting What You Pay For? Auditing Model Substitution in LLM APIs [71.7892165868749]
  Commercial Large Language Model (LLM) APIs create a fundamental trust problem.<n>Users pay for specific models but have no guarantee that providers deliver them faithfully.<n>We formalize this model substitution problem and evaluate detection methods under realistic adversarial conditions.<n>We propose and evaluate the use of Trusted Execution Environments (TEEs) as one practical and robust solution.
  arXiv  Detail & Related papers  (2025-04-07T03:57:41Z)
• Conservative Prediction via Data-Driven Confidence Minimization [70.93946578046003]
  In safety-critical applications of machine learning, it is often desirable for a model to be conservative. We propose the Data-Driven Confidence Minimization framework, which minimizes confidence on an uncertainty dataset.
  arXiv  Detail & Related papers  (2023-06-08T07:05:36Z)
• RobustBench: a standardized adversarial robustness benchmark [84.50044645539305]
  Key challenge in benchmarking robustness is that its evaluation is often error-prone leading to robustness overestimation. We evaluate adversarial robustness with AutoAttack, an ensemble of white- and black-box attacks. We analyze the impact of robustness on the performance on distribution shifts, calibration, out-of-distribution detection, fairness, privacy leakage, smoothness, and transferability.
  arXiv  Detail & Related papers  (2020-10-19T17:06:18Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.