The Mask Is Not the Model: Auditing Prefix Invariance in Attention, State-Space, and Hybrid Sequence Models
A prefix-invariance audit detects future-token leakage across attention, state-space, and hybrid sequence models without training or gradients.
Excerpt
Taebong Kim, Youngsik Hong, Minsik Kim, Sunyoung Choi, Jaewon Jang — We formalize prefix invariance: representations at position t must not depend on future inputs. We give a lightweight audit, two forward passes, no training or gradients, that localizes exactly where causality breaks. Attention-mask inspection is incomplete: leaks can occur via scans or normalization despite correct masks. Across 192 injected-fault trials on eight checkpoints, mask inspection found none, while our audit localized all 192/192, also finding a defect in Zamba2 and Nemotron-H.
Read at source: https://arxiv.org/abs/2608.22876