# Cross-Family Universality of Behavioral Axes via Anchor-Projected Representations

Source: [Cross-Family Universality of Behavioral Axes via Anchor-Projected Representations](<https://www.sota2.com/research/paper/6a02b48edbc836b87d824cd0>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-05-13T05:00:20.520Z

Authors: Su-Hyeon Kim, Yo-Sub Han

Published: 2026-05-11T02:01:18.000Z

## Abstract

Large language models from different families use different hidden dimensions, tokenizers, and training procedures, making behavioral directions difficult to compare or transfer across models. We introduce an anchor-projection framework that maps hidden representations from each model into a shared anchor coordinate space (ACS). Behavioral directions extracted from source models are projected into ACS and averaged into a canonical direction. For a new model, the canonical direction is reconstructed into its native hidden space using only anchor activations, without fine-tuning or target-specific direction extraction. We evaluate five instruction-tuned model families and ten behavioral axes. We find that same-axis directions align tightly across the Llama-Qwen-Mistral-Phi (LQMP) cluster in ACS. This shared structure transfers to downstream tasks. For the aligned LQMP cluster, held-out targets achieve (0.83) ten-way detection accuracy and (0.95) mean binary AUROC, while canonical steering induces refusal-rate shifts of up to +0.46% under distribution shift. Sensitivity analyses show that two source models and small anchor pools already suffice to approximate transferable directions. Overall, ACS provides a novel perspective on cross-family interpretability, revealing that representation-level transfer remains robust across model families.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Behavioral Axis Detection on ACS In-distribution 10 axes (test)](<https://www.sota2.com/research/sota/behavioral-axis-detection-on-acs-in-distribution-10-axes-test.md>): Behavioral Axis Detection; ACS In-distribution 10 axes (test); Refusal MC Recall: 97.

- [Refusal Detection on JailbreakBench Refusal OOD (test)](<https://www.sota2.com/research/sota/refusal-detection-on-jailbreakbench-refusal-ood-test.md>): Refusal Detection; JailbreakBench Refusal OOD (test); MC Recall: 100.

- [Refusal Detection on XSTest Refusal OOD (test)](<https://www.sota2.com/research/sota/refusal-detection-on-xstest-refusal-ood-test.md>): Refusal Detection; XSTest Refusal OOD (test); MC Recall: 100.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2605.09875>)

[PDF](<https://arxiv.org/pdf/2605.09875.pdf>)