Revealing Hidden Model Behaviors with Task-Specific Self-Reports

• URL: http://arxiv.org/abs/2607.03640v1
• Date: Fri, 03 Jul 2026 23:41:25 GMT
• Title: Revealing Hidden Model Behaviors with Task-Specific Self-Reports
• Abstract summary: We introduce the Stabilized Adapter for self-Report (SAR), a lightweight LoRA adapter that makes a fine-tuned model describe its own hidden behavior in plain language.<n>Across seven implanted behaviors (plus a no-behavior control), SAR detects the hidden behavior in every one--even when the model has generalized into broad misalignment.
• Score: 0.0
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Fine-tuning can give a language model a hidden behavior--it may give false answers under a narrow condition, or give harmful advice only when a prompt touches a particular topic. We introduce the Stabilized Adapter for self-Report (SAR), a lightweight LoRA adapter that makes a fine-tuned model describe its own hidden behavior in plain language, using only the model and the dataset it was trained on. Across seven implanted behaviors (plus a no-behavior control), SAR detects the hidden behavior in every one--even when the model has generalized into broad misalignment that the training data alone does not predict. Introspection Adapters (IA), the closest existing baseline, detects some behaviors from our suite but misses others entirely--and where it misses, it hallucinates, consistently reporting wrong behaviors. SAR retains positive signal on every setting where IA fails and halves the rate of hallucinations. This makes it much easier for practitioners to audit their models and obtain reliable answers to "what did my model actually learn?" type of questions.

Related papers

• Your LLM, Your Style: Behavioral Mode Axes for LLM Behavioral Control [15.236237731175192]
  Large language models (LLMs) increasingly act in interactive settings where their behavioral styles affect user experience, safety, and downstream decision making.<n>Existing LLM personality studies largely rely on self-report questionnaires administered in first-person settings.<n>We introduce a situated behavioral-data framework for studying and controlling LLM behavioral personality.
  arXiv  Detail & Related papers  (2026-08-11T09:25:47Z)
• Do Linear Probes Generalize Better in Persona Coordinates? [0.0]
  We study whether there exists a low-dimensional subspace of the model internals that captures harmful behaviors more robustly.<n>We construct persona axes for deception and sycophancy using contrastive persona prompts.<n>We show that persona-derived directions transfer non-trivially and probes trained on persona-PC projections generalize better than probes trained on raw activations.
  arXiv  Detail & Related papers  (2026-05-10T07:38:34Z)
• Split Personality Training: Revealing Latent Knowledge Through Alternate Personalities [15.59200865541989]
  We introduce Split Personality Training (SPT) to fine-tune a second honest persona'' into parameters that remain inactive during normal operation.<n>SPT achieves 96% overall accuracy, whereas Anthropic reports near 0% accuracy.
  arXiv  Detail & Related papers  (2026-02-05T10:45:48Z)
• From Data to Behavior: Predicting Unintended Model Behaviors Before Training [78.37660873165284]
  We introduce Data2Behavior, a new task for predicting unintended model behaviors prior to training.<n>We also propose Manipulating Data Features (MDF), a lightweight approach that summarizes candidate data through their mean representations.<n>Experiments on Qwen3-14B, Qwen2.5-32B-Instruct, and Gemma-3-12b-it confirm that MDF can anticipate unintended behaviors and provide insight into pre-training vulnerabilities.
  arXiv  Detail & Related papers  (2026-02-04T16:37:17Z)
• The 'Sure' Trap: Multi-Scale Poisoning Analysis of Stealthy Compliance-Only Backdoors in Fine-Tuned Large Language Models [10.377264470934843]
  Backdoor attacks on large language models typically couple a secret trigger to an explicit malicious output.<n>We introduce a compliance-only backdoor: supervised fine-tuning on a mostly benign dataset in which a small subset of prompts is suffixed with an arbitrary single-word trigger.<n>We conduct a multi-scale analysis of this benign-label poisoning behavior across poison budget, total fine-tuning dataset size, and model size.
  arXiv  Detail & Related papers  (2025-11-16T02:01:58Z)
• Probing for Arithmetic Errors in Language Models [86.8227317662622]
  Internal activations in language models can be used to detect arithmetic errors.<n>We show that simple probes can accurately decode both the model's predicted output and the correct answer from hidden states.<n>We train lightweight error detectors that predict model correctness with over 90% accuracy.
  arXiv  Detail & Related papers  (2025-07-16T16:27:50Z)
• Auditing language models for hidden objectives [26.85568746300155]
  We study the feasibility of conducting alignment audits: investigations into whether models have undesired objectives.<n>As a testbed, we train a language model with a hidden objective. Our training pipeline first teaches the model about exploitable errors in RLHF reward models (RMs), then trains the model to exploit some of these errors.<n>We leverage this model to study alignment audits in two ways. First, we conduct a blind auditing game where four teams, unaware of the model's hidden objective or training, investigate it for concerning behaviors and their causes.<n>Three teams successfully uncovered the model's hidden objective using techniques including
  arXiv  Detail & Related papers  (2025-03-14T00:21:15Z)
• Eliciting Language Model Behaviors with Investigator Agents [93.34072434845162]
  Language models exhibit complex, diverse behaviors when prompted with free-form text.<n>We study the problem of behavior elicitation, where the goal is to search for prompts that induce specific target behaviors.<n>We train investigator models to map randomly-chosen target behaviors to a diverse distribution of outputs that elicit them.
  arXiv  Detail & Related papers  (2025-02-03T10:52:44Z)
• Unfamiliar Finetuning Examples Control How Language Models Hallucinate [75.03210107477157]
  Large language models are known to hallucinate when faced with unfamiliar queries. We find that unfamiliar examples in the models' finetuning data are crucial in shaping these errors. Our work further investigates RL finetuning strategies for improving the factuality of long-form model generations.
  arXiv  Detail & Related papers  (2024-03-08T18:28:13Z)
• GPT-HateCheck: Can LLMs Write Better Functional Tests for Hate Speech Detection? [50.53312866647302]
  HateCheck is a suite for testing fine-grained model functionalities on synthesized data. We propose GPT-HateCheck, a framework to generate more diverse and realistic functional tests from scratch. Crowd-sourced annotation demonstrates that the generated test cases are of high quality.
  arXiv  Detail & Related papers  (2024-02-23T10:02:01Z)
• Probing Model Signal-Awareness via Prediction-Preserving Input Minimization [67.62847721118142]
  We evaluate models' ability to capture the correct vulnerability signals to produce their predictions. We measure the signal awareness of models using a new metric we propose- Signal-aware Recall (SAR) The results show a sharp drop in the model's Recall from the high 90s to sub-60s with the new metric.
  arXiv  Detail & Related papers  (2020-11-25T20:05:23Z)

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.