Provenance Generalization and Verification Scaling for Neural Network Forensics

Abstract

Prior work established that knowledge distillation transfers a detectable provenance trace from teacher to student models, and that API endpoint verification can identify models through logprob order-statistic geometry. Both results were demonstrated on single teacher-student pairs and a six-model API zoo, leaving open whether provenance detection generalizes across model families and whether API verification scales to production-density endpoint populations. We address both questions through a coordinated experimental program spanning four studies. In the first study, we train 24 distilled checkpoints across 7 experimental arms — 3 teacher families (Qwen, Mistral, Llama), 4 student architectures (Qwen-0.5B, Qwen-1.5B, Llama-1B, Gemma-2B), and 2 training protocols (logit-level knowledge distillation and cross-tokenizer supervised fine-tuning) — measuring provenance transfer in both the weight-geometry and API-logprob regimes. Provenance transfer generalizes across the tested matrix: all 14 mature-epoch checkpoints show directional coupling to the teacher (cosine alignment cosθ > 0.8, with 13 of 14 exceeding 0.85). The strongest signal arises in a cross-family arm (Mistral-7B → Llama-1B, scalar convergence 0.858) that is inconsistent with a purely family-restricted transfer hypothesis within the tested matrix. The normalized third logit gap δ_norm remains within 1.4% coefficient of variation across all 31 checkpoints and 4 student architectures — the tightest confirmation of Gumbel-class universality in this experimental program. An extension to mixture-of-experts architecture (Mixtral-8x7B, δ_norm = 0.309) confirms that the universal constant persists under sparse expert routing. In the second contribution, we identify a systematic failure mode of scalar provenance metrics and introduce the geometrically correct directional diagnostic for provenance detection in inner-product spaces. The standard scalar convergence metric Conv_T conflates direction and magnitude into a single value, discarding the directional information that provenance detection requires. In two independent experiments, this produced misleading conclusions: a false spoofing signal (R^2 = 0.995 of apparent cross-family convergence explained by pure knowledge distillation geometry, with the adversarial gradient contributing 4.8%) and a false failure signal (negative Conv_T despite consistent directional coupling at cosθ = 0.91). The alignment diagnostic applies the law of cosines in PPP-residual template space (vectors in R^K with Euclidean distance) to decompose student movement into direction and magnitude, preserving the provenance signal that scalar distance metrics destroy. We establish a measurability threshold: when the baseline-to-teacher distance d(B,T) falls below approximately 1.0, scalar Conv_T becomes unreliable and the directional diagnostic becomes the primary metric. This diagnostic applies to any distillation forensics framework that measures convergence in an inner-product space. In the third contribution, we extend API endpoint verification from 6 models to 14 across 3 commercial providers (OpenAI, Google Vertex AI, xAI), observing zero breaches across 182 pairwise impostor comparisons under per-model adaptive thresholds and three independent enrollment sessions, with a centroid reference protocol (CRP) that replaces the centroid L^2 metric, which produces false breaches at 14-model density. We establish a minimum truncation floor: API endpoints exposing fewer than 7 logprob ranks cannot support reliable verification (signal collapses within one rank of this boundary). Speculative decoding — an increasingly common inference optimization — is shown to be transparent to the verification protocol, with the speculative-decoded fingerprint deviating from the verifier-only fingerprint by 10.6% of the inter-model distance. Finally, we formalize the Trust Paradox in model forensics — a victim cannot prove weight theft without disclosing weights, and a suspect cannot prove innocence without disclosing training data — and propose a three-tier zero-knowledge attestation architecture that addresses it. The first tier (committed distance proof) enables a model owner to prove fingerprint proximity to a public anchor without revealing the fingerprint vector, using standard cryptographic commitments with verifier-controlled thresholds. The second tier (hardware-attested measurement) removes the requirement that the prover be trusted to compute the fingerprint correctly, binding the measurement to a trusted execution environment attestation. The third tier (full zero-knowledge extraction) would eliminate all trust assumptions beyond cryptographic soundness; we present this as an open problem with pre-registered falsification criteria, including a fixed-point precision gate derived from the minimum pairwise separation in the existing 23-model zoo. The architecture defines eight properties that a meaningful zero-knowledge model identity proof must satisfy — extending the formal verification doctrine (311 + 41 = 352 theorems across 17 Coq proof files [1, 2], 0 Admitted) into the cryptographic regime — and six explicit trust assumptions under which the proof statements hold. All three tiers are validated: Tier 1 (committed distance proof) has been implemented and hardened; Tier 2 (hardware-attested measurement) has been validated on production confidential computing hardware (6 models, 1,536 measurements, 0 failures inside an H100 trusted execution environment, with both CPU and GPU attestation tokens bound to a common cryptographic root and structural fingerprints transparent to confidential computing mode); and Tier 3 (full zero-knowledge extraction) has been validated — a complete circuit has been compiled and audited, all four pre-registered falsification criteria have been met, and the proof system operates within practical proving-time and proof-size bounds. The breakthrough discoveries enabled by Tier 3 validation, including an identity-conditioned inference verification architecture, are reported in the companion paper. The experimental results in this paper are grounded in the formal verification stack and measurement infrastructure described in the companion papers [1, 2, 3]. All provenance claims are classified as VALIDATED (empirical); Tier 1 (committed distance proof) has been implemented and hardened, and Tier 2 (hardware-attested measurement) has been validated on production confidential computing hardware — both are classified VALIDATED. Tier 3 (full zero-knowledge extraction) has been validated: a complete circuit was compiled and audited, all four pre-registered falsification criteria were met, and the architecture has been extended into identity-conditioned inference verification [6]. The Neural Network Identity Series — Mathematical foundations, empirical validation, and governance frameworks for verifying which model is running Newest addition: Technical Note: The Disappearing Window — AI Logprob Access Withdrawal and the Structural Verifiability of Frontier Model Contracts (DOI: 10.5281/zenodo.20362098) Paper 1: The δ-Gene: Inference-Time Physical Unclonable Functions from Architecture-Invariant Output Geometry (DOI: 10.5281/zenodo.18704275) Paper 2: Template-Based Endpoint Verification via Logprob Order-Statistic Geometry (DOI: 10.5281/zenodo.18776711) Paper 3: The Geometry of Model Theft: Distillation Forensics, Adversarial Erasure, and the Illusion of Spoofing (DOI: 10.5281/zenodo.18818608) Paper 4: Provenance Generalization and Verification Scaling for Neural Network Forensics (DOI: 10.5281/zenodo.18872071) Paper 5: Beneath the Character: The Structural Identity of Neural Networks — Mathematical Evidence for a Non-Narrative Layer of AI Identity (DOI: 10.5281/zenodo.18907292) Paper 6: Which Model Is Running?: Structural Identity as a Prerequisite for Trustworthy Zero-Knowledge Machine Learning (DOI: 10.5281/zenodo.19008116) Paper 7: The Deformation Laws of Neural Identity (DOI: 10.5281/zenodo.19055966) Paper 8: What Counts as Proof? — Admissible Evidence for Neural Network Identity Claims (DOI: 10.5281/zenodo.19058540) Paper 9: Composable Model Identity — Formal Hardening of Structural Attestations in the Enterprise Identity Stack (DOI: 10.5281/zenodo.19099911) Paper 10:Where Identity Comes From: Path Sensitivity and Endpoint Underdetermination in Neural Network Training (DOI: 10.5281/zenodo.19118807) Paper 11: Post-Hoc Disclosure Is Not Runtime Proof: Model Identity at Frontier Scale (DOI: 10.5281/zenodo.19216634) Paper 12: Family-Dependent Response to Reasoning Distillation Across Structural and Functional Identity Layers (DOI: 10.5281/zenodo.19298857) Paper 13: Safety-Alignment Removal as a Model-Identity Failure — Structural Evidence from Published Weight-Level Mutation Checkpoints (DOI: 10.5281/zenodo.19383019) Technical Note: Agent Identity Is Not Model Identity (DOI: 10.5281/zenodo.19240883) Technical Note: Gap Invariance: Why PPP Measurements Are Domain-Independent by Construction (DOI: 10.5281/zenodo.19275524) Technical Note: Measured Model Substitution Under Valid Agent Credentials (DOI: 10.5281/zenodo.19342848) Technical Note: Artifact Identity Is Not Runtime Identity — Trustfall Lite and the Boundary of File-Level Model Verification (DOI: 10.5281/zenodo.20019127) Formal Verification Stack for Neural Network Structural Identity (IT-PUF Coq Proofs) (DOI: 10.5281/zenodo.18930621) Copyright (c) 2026 Anthony Ray Coslett / Fall Risk AI, LLC. All Rights Reserved. Confidential and Proprietary. Patent Pending (Applications 63/982,893, 63/990,487, 63/996,680, 64/003,244).