Search papers, labs, and topics across Lattice.
This study investigates the impact of fine-tuning on embedded activation steering in language models, specifically focusing on refusal suppression and brevity induction across various instruction-tuned models. The findings reveal that while the steering mechanism remains intact during fine-tuning, its behavioral effects can significantly degrade, losing an average of 64% effectiveness under supervised fine-tuning (SFT). This highlights a critical distinction between the durability of the weight edits and the functional vulnerability of the behaviors they induce, necessitating re-validation of model behavior post-training.
Fine-tuning may preserve the underlying steering mechanism, but it can drastically undermine the intended behavioral effects, with an average 64% loss in effectiveness.
Activation steering can be embedded directly into a language model's weights, shaping behaviour without inference-time intervention and offering a way to encode alignment prior to release. However, models are routinely fine-tuned after deployment, and it is unknown whether embedded interventions survive this. We study the stability of embedded steering for refusal suppression and brevity induction across five instruction-tuned models (3B-14B) under non-adversarial SFT and RLHF. Behaviourally, preservation tracks the training data: steering degrades when optimisation pressure contradicts the targeted behaviour and persists otherwise, with refusal ablation losing 64% of its effect on average under SFT. Mechanistically, however, the weight edit survives almost untouched even where behaviour reverts: mean vector recovery is $\rho = 0.004$, and the fine-tuning update along the steering direction is near-orthogonal to its pre-edit weight pattern (mean $\cos\theta = 0.074$). When steered behaviour degrades, fine-tuning does not achieve it by dismantling or reversing the steering mechanism itself. Embedded steering is therefore mechanistically durable but functionally vulnerable, and requires behavioural re-validation after downstream training.
