Search papers, labs, and topics across Lattice.
This paper introduces an unsupervised method to detect and mitigate subliminal behavioral traits inherited by distilled language models, even when the trait is transferred via semantically unrelated data and its nature is unknown beforehand. The technique leverages the consistent expression of the trait across diverse teacher-generated datasets, contrasting with the variable nature of task-specific capabilities. By identifying a weight-space direction associated with the trait, the method can both reveal and suppress these inherited misalignments with minimal impact on downstream performance.
Unsupervisedly uncover and neutralize hidden model misalignments transferred during distillation, even without prior knowledge of the problematic trait.
When a language model is distilled from a teacher that shares its base parameters, the student can inherit behavioural traits of the teacher---including misalignment---even from data that is semantically unrelated to the trait (Cloud et al., 2026). All existing defences require knowing the trait. We remove this requirement and prove that a subliminally transferred trait can be recovered from the parameter updates alone, without supervision, by exploiting a single asymmetry: the trait is expressed consistently across diverse datasets generated by the teacher, whereas task-specific and shared capabilities vary. The resulting estimator returns a weight-space direction that can be amplified to discover and interpret the trait or projected out to suppress it. Applied to the subliminal-learning benchmarks of (Cloud et al., 2026), our method recovers the transmitted animal-preference traits without any trait labels and suppresses their expression with almost no impact to task performance.
