From Data to Behavior: Predicting Unintended Model Behaviors Before Training

ACL ARR 2026 January Submission1730 Authors

Keywords: unintended behavior prediction, subliminal learning, before training, model mechanism

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.

Paper Type: Long

Research Area: Safety and Alignment in LLMs

Research Area Keywords: alignment, interpretability, bias

Contribution Types: Model analysis & interpretability

Languages Studied: English

Submission Number: 1730