Latent-space Attacks for Refusal Evasion in Language Models

ArXi:2605.21706v1 Announce Type: new Safety-aligned language models are trained to refuse harmful requests, yet refusal behavior can be suppressed by steering their internal representations. Existing methods do so by ablating a refusal direction from model activations, aiming to remove refusal from the model's residual stream. Despite their empirical success, these methods lack a principled account of the latent-space transformation they induce and why it suppresses refusal.