arXiv cs.AI· Pierre Peign\'e, Ky Nguyen, Paul Wang·· 13 小时前AI 评分31
前沿 AI 训练零知识验证的提案与挑战
Proposals and challenges for zero-knowledge verification of frontier AI training
AI 导读
针对前沿 AI 训练算力治理，研究者提出一种结合预承诺规范、节点间网络观测与中间计算实时 Merkle 承诺的密集预训练设计，并给出带原生 BF16/FP32 预编译的参考 zkVM，可在被挑战位置逐位核验已承诺计算。
正文 · 原文
Abstract:Proposals for Frontier AI rules often use cumulative training compute as the main criterion for identifying high-impact models. Recent governance analyses call zero-knowledge proofs promising but currently impractical at frontier scale [26, 4]. We argue this barrier reflects the current paradigm, not a fundamental limit. Our frontier dense-pre-training design combines a pre-committed specification, inter-node network observations, and live Merkle commitments to intermediate computation. Our reference zero-knowledge virtual machine (zkVM) provides a succinct proof layer with native BF16/FP32 precompiles. We qualitatively compare zkVMs, custom circuits, and direct interactive audits by transferability, soundness, and cost. Public randomness beacons are a separate audit-selection choice.
At challenged positions, proofs check the committed computation against declared BF16/FP32 semantics bit-for-bit. A private training specification targets model-architecture confidentiality. Genesis proofs cover initialisation; then, in-training step proofs verify selected steps across the run. Ex-ante attestations enforce policy-relevant claims as running invariants on every proved step. Together they create a governance-auditable training record. Detection-probability formulas quantify sampling guarantees from the sampling budget and the share of computation altered. We give numerical examples and explain what sampling does not cover.
We note, however, that weight smuggling could reuse or merge weights trained elsewhere without charging that work to the current run. We then argue that regimes with global training budgets are more robust than regimes with per-run FLOP caps. This advantage depends on counting all contributing training. We catalogue remaining research and engineering problems for external contributors.
| Comments: | 71 pages, 2 figures |
| Subjects: | Artificial Intelligence (cs.AI); Systems and Control (eess.SY) |
| Cite as: | arXiv:2606.05433 [cs.AI] |
| (or arXiv:2606.05433v3 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2606.05433 arXiv-issued DOI via DataCite |
Submission history
From: Paul Wang [view email]
[v1]
Wed, 3 Jun 2026 20:57:28 UTC (380 KB)
[v2]
Sat, 22 Aug 2026 05:46:47 UTC (381 KB)
[v3]
Wed, 21 Oct 2026 16:10:51 UTC (426 KB)
来源：arXiv cs.AI · arxiv.org