When State Space Models Forget: A New Attack and Its Defense
![Under adversarial attack utilizing HiSPA, the Mamba-130M model experiences a precipitous decline in information retention-falling from [latex]\rho(\bar{A}) = 0.98[/latex] to [latex]0.32[/latex]-resulting in a [latex]52.5[/latex] percentage-point accuracy collapse, manifested as a contraction of hidden-state trajectories toward the origin, though this collapse can be mitigated by SpectralGuard intervention.](https://arxiv.org/html/2603.12414v1/figures/hidden_state_trajectories_3d.png)
Researchers have discovered a vulnerability in State Space Models where adversarial attacks can cause a ‘spectral radius collapse’, leading to performance degradation, and propose a novel monitoring system to detect and mitigate these attacks.


![The system demonstrates effective tail packing within the cipher of [latex]r\_0^\widehat{\bm{r}\_{0}}[/latex], achieved through strategic management of in-queue inputs.](https://arxiv.org/html/2603.12946v1/x5.png)


![The system explores the inherent limitations of multi-UAV collision avoidance, demonstrating that successful navigation depends not simply on finding <i>a</i> solution-a control input within admissible bounds-but on the <i>existence</i> of a consistent intersection among all collision-avoidance constraints [latex] \bigcap_{i} C_i \neq \emptyset [/latex], a condition termed ‘internal compatibility’ which precedes any consideration of feasible control actions.](https://arxiv.org/html/2603.13103v1/x1.png)

