LymphNode: A Plug-and-Play Access Control Method for Deep Neural Networks

Deep Neural Networks (DNNs) are high-value intellectual property (IP), yet deploying them to edge environments exposes them to $\mathbf{\text{unrestricted}}\mathbf{\text{oracle}}\mathbf{\text{access}}$, rendering them vulnerable to model extraction and inversion attacks. Existing defenses fail to address this practically: passive watermarking only offers post-hoc provenance, while active defenses impose prohibitive latency or require persistent access to sensitive training data. To bridge this gap, we propose $\mathit{\text{LymphNode}}$, a novel post-hoc defense framework that acts as an intrinsic immune system" within the model. \textit{LymphNode} enforces a strict default-deny’’ policy: it actively neutralizes model utility for unauthorized queries via $\mathbf{\text{Generalized}}\mathbf{\text{Sparse}}\mathbf{\text{Universal}}\mathbf{\text{Adversarial}}\mathbf{\text{Perturbations}}\mathbf{\text{(GSUAP)}}$ injected into the feature space, effectively blocking gradient estimation and data inference. Utility is selectively restored only for authorized inputs carrying a stealthy feature-domain credential. Our framework is highly practical: it is $\mathbf{\text{data-efficient}}$, establishing robust protection with fewer than 100 samples ($<1\mathrm{\%}$ of training data), and $\mathbf{\text{cross-dataset}}\mathbf{\text{adaptable}}$, enabling protection using public surrogate datasets. $\mathit{\text{LymphNode}}$ thus provides a lightweight, immediately deployable defense for high-stakes scenarios where original training data is restricted or unavailable.