From Data to Behavior: Predicting Unintended Model Behaviors Before Training

Mengru Wang, Zhenqian Xu, Junfeng Fang, Yunzhi Yao, Shumin Deng +2 more
2/4/2026
cs.LGcs.AIcs.CL

No official code found. Best implementation: zjunlp/Data2Behavior (5 stars, MIT, last commit 2026-02-04).

Abstract

Large Language Models (LLMs) can acquire unintended biases from seemingly benign training data even without explicit cues or malicious content. Existing methods struggle to detect such risks before fine-tuning, making post hoc evaluation costly and inefficient. To address this challenge, we introduce Data2Behavior, a new task for predicting unintended model behaviors prior to training. We also propose Manipulating Data Features (MDF), a lightweight approach that summarizes candidate data through their mean representations and injects them into the forward pass of a base model, allowing latent statistical signals in the data to shape model activations and reveal potential biases and safety risks without updating any parameters. MDF achieves reliable prediction while consuming only about 20% of the GPU resources required for fine-tuning. 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.

View on arXivView PDF

Code Implementations(1)

zjunlp/Data2BehaviorCommunity72%
MIT

From Data to Behavior: Predicting Unintended Model Behaviors Before Training

50Dec 22, 20258 months agoMIT
data2behaviorlarge-language-modellarge-language-modelsnatural-language-processingsteering-behaviors+1 more

Similar Papers

In-the-Wild Model Organisms: Mitigating Undesirable Emergent Behaviors in Production LLM Post-Training via Data Attribution

86% match2026

Making Harmful Behaviors Unlearnable for Large Language Models

86% match2023

Language Models Can Predict Their Own Behavior

84% match1 repo22025

Where Did It Go Wrong? Attributing Undesirable LLM Behaviors via Representation Gradient Tracing

84% match1 repo52025

In-Training Defenses against Emergent Misalignment in Language Models

84% match2025

DebugLM: Learning Traceable Training Data Provenance for LLMs

84% match2026

Cite this paper

@article{wang2026from,
  title  = {From Data to Behavior: Predicting Unintended Model Behaviors Before Training},
  author = {Mengru Wang and Zhenqian Xu and Junfeng Fang and Yunzhi Yao and Shumin Deng and Huajun Chen and Ningyu Zhang},
  year   = {2026},
  eprint = {2602.04735},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2602.04735v1}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2602.04735.svg)](https://codeofpaper.com/paper/2602.04735)

Discussion