The Assistant Axis: Situating and Stabilizing the Default Persona of Language Models

Christina Lu, Jack Gallagher, Jonathan Michala, Kyle Fish, Jack Lindsey
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:80955-81003, 2026.

Abstract

Large language models can represent a variety of personas but typically default to a helpful Assistant identity cultivated during post-training. Across several different models, we find an “Assistant Axis" in their activation space, which captures the extent to which a model is operating in its default Assistant mode. Steering towards the Assistant direction reinforces helpful and harmless behavior; steering away increases the model’s tendency to identify as other entities. Measuring deviations along the Assistant Axis predicts “persona drift,” a phenomenon where models slip into exhibiting harmful or bizarre behaviors that are uncharacteristic of their typical persona. We find that persona drift is often driven by conversations demanding meta-reflection on the model’s processes or featuring emotionally vulnerable users. We show that restricting activations to a fixed region along the Assistant Axis can stabilize model behavior in these scenarios—and also in the face of adversarial persona-based jailbreaks. Our results suggest that post-training steers models toward a particular region of persona space but only loosely tethers them to it, motivating work on training and steering strategies that more deeply anchor models to a coherent persona.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-lu26g, title = {The Assistant Axis: Situating and Stabilizing the Default Persona of Language Models}, author = {Lu, Christina and Gallagher, Jack and Michala, Jonathan and Fish, Kyle and Lindsey, Jack}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {80955--81003}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/lu26g/lu26g.pdf}, url = {https://proceedings.mlr.press/v306/lu26g.html}, abstract = {Large language models can represent a variety of personas but typically default to a helpful Assistant identity cultivated during post-training. Across several different models, we find an “Assistant Axis" in their activation space, which captures the extent to which a model is operating in its default Assistant mode. Steering towards the Assistant direction reinforces helpful and harmless behavior; steering away increases the model’s tendency to identify as other entities. Measuring deviations along the Assistant Axis predicts “persona drift,” a phenomenon where models slip into exhibiting harmful or bizarre behaviors that are uncharacteristic of their typical persona. We find that persona drift is often driven by conversations demanding meta-reflection on the model’s processes or featuring emotionally vulnerable users. We show that restricting activations to a fixed region along the Assistant Axis can stabilize model behavior in these scenarios—and also in the face of adversarial persona-based jailbreaks. Our results suggest that post-training steers models toward a particular region of persona space but only loosely tethers them to it, motivating work on training and steering strategies that more deeply anchor models to a coherent persona.} }
Endnote
%0 Conference Paper %T The Assistant Axis: Situating and Stabilizing the Default Persona of Language Models %A Christina Lu %A Jack Gallagher %A Jonathan Michala %A Kyle Fish %A Jack Lindsey %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-lu26g %I PMLR %P 80955--81003 %U https://proceedings.mlr.press/v306/lu26g.html %V 306 %X Large language models can represent a variety of personas but typically default to a helpful Assistant identity cultivated during post-training. Across several different models, we find an “Assistant Axis" in their activation space, which captures the extent to which a model is operating in its default Assistant mode. Steering towards the Assistant direction reinforces helpful and harmless behavior; steering away increases the model’s tendency to identify as other entities. Measuring deviations along the Assistant Axis predicts “persona drift,” a phenomenon where models slip into exhibiting harmful or bizarre behaviors that are uncharacteristic of their typical persona. We find that persona drift is often driven by conversations demanding meta-reflection on the model’s processes or featuring emotionally vulnerable users. We show that restricting activations to a fixed region along the Assistant Axis can stabilize model behavior in these scenarios—and also in the face of adversarial persona-based jailbreaks. Our results suggest that post-training steers models toward a particular region of persona space but only loosely tethers them to it, motivating work on training and steering strategies that more deeply anchor models to a coherent persona.
APA
Lu, C., Gallagher, J., Michala, J., Fish, K. & Lindsey, J.. (2026). The Assistant Axis: Situating and Stabilizing the Default Persona of Language Models. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:80955-81003 Available from https://proceedings.mlr.press/v306/lu26g.html.

Related Material