{"paper":{"title":"SpanKey: Dynamic Key Space Conditioning for Neural Network Access Control","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"SpanKey gates neural network inference by conditioning activations on keys from a defined low-dimensional subspace.","cross_cats":["cs.AI"],"primary_cat":"cs.CR","authors_text":"WenBin Yan","submitted_at":"2026-04-14T04:01:34Z","abstract_excerpt":"SpanKey is a lightweight way to gate inference without encrypting weights or chasing leaderboard accuracy on gated inference. The idea is to condition activations on secret keys. A basis matrix $B$ defines a low-dimensional key subspace $Span(B)$; during training we sample coefficients $\\alpha$ and form keys $k=\\alpha^\\top B$, then inject them into intermediate activations with additive or multiplicative maps and strength $\\gamma$. Valid keys lie in $Span(B)$; invalid keys are sampled outside that subspace. We make three points. (i) Mechanism: subspace key injection and a multi-layer design sp"},"claims":{"count":4,"items":[{"kind":"strongest_claim","text":"Subspace key injection with multi-layer design, together with deny losses and margin-tail diagnostics, enables practical key-based gating of neural network inference, as demonstrated by CIFAR-10 ResNet-18 runs and MNIST ablations.","source":"verdict.strongest_claim","status":"machine_extracted","claim_id":"C1","attestation":"unclaimed"},{"kind":"weakest_assumption","text":"That the network does not absorb the key signal into its weights in a way that collapses separation between valid and invalid keys at deployment scale, despite the analytical Beta-energy split and margin diagnostics provided.","source":"verdict.weakest_assumption","status":"machine_extracted","claim_id":"C2","attestation":"unclaimed"},{"kind":"one_line_summary","text":"SpanKey injects keys from a learned subspace into network activations via additive or multiplicative maps to enable key-based access control for neural network inference.","source":"verdict.one_line_summary","status":"machine_extracted","claim_id":"C3","attestation":"unclaimed"},{"kind":"headline","text":"SpanKey gates neural network inference by conditioning activations on keys from a defined low-dimensional subspace.","source":"verdict.pith_extraction.headline","status":"machine_extracted","claim_id":"C4","attestation":"unclaimed"}],"snapshot_sha256":"62f76804eabf2e507acf835b0c3ca516278b181c8ecc5b9be102a528ec5c21e7"},"source":{"id":"2604.12254","kind":"arxiv","version":2},"verdict":{"id":"1bb38d14-2076-4f4e-b1b0-c0babc128d08","model_set":{"reader":"grok-4.3"},"created_at":"2026-05-10T15:54:32.601623Z","strongest_claim":"Subspace key injection with multi-layer design, together with deny losses and margin-tail diagnostics, enables practical key-based gating of neural network inference, as demonstrated by CIFAR-10 ResNet-18 runs and MNIST ablations.","one_line_summary":"SpanKey injects keys from a learned subspace into network activations via additive or multiplicative maps to enable key-based access control for neural network inference.","pipeline_version":"pith-pipeline@v0.9.0","weakest_assumption":"That the network does not absorb the key signal into its weights in a way that collapses separation between valid and invalid keys at deployment scale, despite the analytical Beta-energy split and margin diagnostics provided.","pith_extraction_headline":"SpanKey gates neural network inference by conditioning activations on keys from a defined low-dimensional subspace."},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2604.12254/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"}