Skip to main content

Spring Deadline: Sunday, February 15 at 11:59 pm PT. Click here to apply.

Back to Research
Accepted to Mech Interp @ NeurIPS 2025 (Spotlight)Spotlight

Shared Parameter Subspaces and Cross-Task Linearity in Emergently Misaligned Behavior

Daniel Aarao Reis Arturi, Eric Zhang, Andrew Adrian Ansah

Abstract

Recent work has discovered that large language models can develop broadly misaligned behaviors after being fine-tuned on narrowly harmful datasets, a phenomenon known as emergent misalignment (EM). However, the fundamental mechanisms enabling such harmful generalization across disparate domains remain poorly understood. In this work, we adopt a geometric perspective to study EM and demonstrate that it exhibits a fundamental cross-task linear structure in how harmful behavior is encoded across different datasets. Specifically, we find a strong convergence in EM parameters across tasks, with the fine-tuned weight updates showing relatively high cosine similarities, as well as shared lower-dimensional subspaces as measured by their principal angles and projection overlaps.

Citation

Daniel Aarao Reis Arturi, Eric Zhang, Andrew Adrian Ansah. "Shared Parameter Subspaces and Cross-Task Linearity in Emergently Misaligned Behavior". Accepted to Mech Interp @ NeurIPS 2025 (Spotlight).

Resources

Details

Conference
Accepted to Mech Interp @ NeurIPS 2025 (Spotlight)
Authors
3 authors
Recognition
Spotlight Paper

Publish Your Research

Join Algoverse and work with world-class mentors to publish at top AI conferences.

Start Your Application