Papers
Every paper
All 87 entries in papers.yml, across every cell. PDF means we hold the primary source; Note means someone has actually read it and written down what to distrust.
87 shown
| Paper | Section | Year | Venue | Headline | Note | |
|---|---|---|---|---|---|---|
| DeepProve | Proving inference | 2026 | IACR ePrint 2026/1112 | 174 tok/min | ✓ | ✓ |
| Jolt Atlas | Proving inference | 2026 | arXiv:2602.17452 | 4.3 tok/min | ✓ | ✓ |
| NANOZK | Proving inference | 2026 | arXiv:2603.18046 | 43 s to prove | ||
| Range-Arithmetic | Proving inference | 2026 | arXiv:2505.17623v2 | — | ✓ | ✓ |
| zkGPT | Proving inference | 2025 | USENIX Security 2025 | 2.75 tok/min | ✓ | ✓ |
| zkPyTorch | Proving inference | 2025 | IACR ePrint 2025/535 | 2.2 s to prove | ✓ | ✓ |
| Bionetta | Proving inference | 2025 | arXiv:2510.06784v2 (technical report, not peer-reviewed) | 3.05 s to prove | ✓ | ✓ |
| ZKTorch | Proving inference | 2025 | arXiv:2507.07031 | 1397.52 s to prove | ✓ | ✓ |
| ZIP | Proving inference | 2025 | ACM CCS 2025 | — | ✓ | ✓ |
| zkLLM | Proving inference | 2024 | ACM CCS 2024 | 900 s to prove | ✓ | ✓ |
| ZKML | Proving inference | 2024 | EuroSys 2024 | 0.016 tok/min | ✓ | ✓ |
| SpaGKR | Proving inference | 2024 | IACR ePrint 2024 | — | ||
| Artemis / Apollo | Proving inference | 2024 | arXiv:2409.12055 | 12000–14400 s to prove | ||
| Lu et al. | Proving inference | 2024 | — | 287.1 s to prove | ||
| Hao et al. | Proving inference | 2024 | USENIX Security 2024 | 2.107 s to prove | ✓ | ✓ |
| Remainder | Proving inference | 2024 | Modulus Labs technical report (not peer-reviewed) | 54 s to prove | ✓ | ✓ |
| ezkl | Proving inference | 2023 | Open-source project (no canonical paper) | 1310 s to prove | ||
| Zator | Proving inference | 2023 | Open-source project, Hack Lodge (no paper) | 26966.8 s to prove | ✓ | |
| zkCNN | Proving inference | 2021 | ACM CCS 2021 | 88.3 s to prove | ||
| Mystique | Proving inference | 2021 | USENIX Security 2021 | 262 s to prove | ✓ | ✓ |
| ZEN | Proving inference | 2021 | — | 147–4710 s to prove | ✓ | ✓ |
| vCNN | Proving inference | 2020 | IEEE TDSC | 28800 s to prove | ||
| SafetyNets | Proving inference | 2017 | NeurIPS 2017 | — | ✓ | ✓ |
| pvCNN | Proving testing | 2023 | — | 1448.83 s to prove | ||
| zkDT | Proving testing | 2020 | ACM CCS 2020 | 250 s to prove | ||
| ZKBoost | Proving training | 2026 | IACR ePrint 2026/202 | — | ||
| Optimum Vicinity | Proving training | 2025 | ACM CCS 2025 | — | ||
| VeriLoRA / zkLoRA | Proving training | 2025 | NDSS 2026 | 600 s to prove | ||
| SUMMER | Proving training | 2025 | IACR ePrint 2025/1688 | — | ||
| ZKPROV | Proving training | 2025 | arXiv:2506.20915 | — | ||
| Kaizen | Proving training | 2024 | ACM CCS 2024 | 900 s to prove | ||
| PoL is More Broken Than You Think | Proving training | 2023 | IEEE EuroS&P 2023 | — | ✓ | ✓ |
| zkDL | Proving training | 2023 | IEEE TIFS 2024 | 0.86 s to prove | ||
| zkPoT (Garg et al.) | Proving training | 2023 | ACM CCS 2023 | 4208 s to prove | ✓ | ✓ |
| PoL Adversarial Examples | Proving training | 2022 | IEEE S&P 2022 | — | ✓ | ✓ |
| zkMLaaS | Proving training | 2022 | IEEE GLOBECOM 2022 | 2.2 s to prove | ||
| Proof-of-Learning | Proving training | 2021 | IEEE S&P 2021 | — | ✓ | ✓ |
| VeriML | Proving training | 2021 | IEEE TPDS | 12 s to prove | ||
| PRoVeFL | Federated & aggregation | 2026 | arXiv:2607.06612 | — | ||
| ByzSFL | Federated & aggregation | 2025 | arXiv:2501.06953 | — | ||
| Trusted Model Aggregation (ZKFL) | Federated & aggregation | 2024 | IEEE TPDS 2024 | — | ||
| RoFL | Federated & aggregation | 2023 | IEEE S&P 2023 | — | ||
| ACORN | Federated & aggregation | 2023 | USENIX Security 2023 | — | ✓ | ✓ |
| zkFL | Federated & aggregation | 2023 | arXiv:2310.02554 | — | ||
| RiseFL | Federated & aggregation | 2023 | — | — | ||
| EIFFeL | Federated & aggregation | 2022 | ACM CCS 2022 | — | ✓ | ✓ |
| Prio / Prio+ | Federated & aggregation | 2017 | NSDI 2017 (Prio); SCN 2022 (Prio+) | — | ||
| Probabilistic truncation in PPML | Numerics | 2025 | AAAI 2025 | — | ✓ | ✓ |
| ZKLP | Numerics | 2024 | arXiv 2404.14983 | — | ✓ | ✓ |
| Garg et al. (FP) | Numerics | 2022 | ACM CCS 2022 | — | ✓ | ✓ |
| SecFloat | Numerics | 2022 | IEEE S&P 2022 | — | ✓ | ✓ |
| Archer et al. (IEEE cost) | Numerics | 2021 | LATINCRYPT 2021 | — | ✓ | ✓ |
| SIRNN | Numerics | 2021 | IEEE S&P 2021 | 49.6 s to prove | ✓ | ✓ |
| Bootstrapping is All You Need | Private inference | 2026 | IACR ePrint 2026/1255 | 349.5 s to infer | ✓ | ✓ |
| Sigma | Private inference | 2024 | PoPETs 2024(4), pp. 61-79 | 37.59 s to prove | ✓ | ✓ |
| BOLT | Private inference | 2024 | IEEE S&P 2024 | — | ✓ | ✓ |
| Nimbus | Private inference | 2024 | NeurIPS 2024 | — | ✓ | ✓ |
| CipherGPT | Private inference | 2023 | IACR ePrint 2023/1147 | 1180.9 s to infer | ✓ | ✓ |
| Cheetah | Private inference | 2022 | USENIX Security 2022 | 80.3 s to prove | ✓ | ✓ |
| Iron | Private inference | 2022 | NeurIPS 2022 | — | ✓ | ✓ |
| Delphi | Private inference | 2020 | USENIX Security 2020 | 3.8 s to prove | ✓ | ✓ |
| Mosformer | Private inference | 2025 | ACM CCS 2025 | 59.47 s to prove | ✓ | ✓ |
| PUMA | Private inference | 2023 | arXiv:2307.12533v3 (preprint, Sep 2023); later published in Security and Safety, 2025 | 200.473 s to prove | ✓ | ✓ |
| PriFT | Private training | 2026 | IACR ePrint 2026/1381 | — | ||
| Private LoRA Fine-tuning with HE | Private training | 2025 | — | — | ||
| CryptPEFT | Private training | 2025 | — | — | ||
| PrivTuner | Private training | 2024 | — | — | ||
| Encryption-Friendly LLM Architecture | Private training | 2024 | — | — | ||
| FairZK | Proving properties | 2025 | IEEE S&P 2025 | 343 s to prove | ||
| Show Me You Comply | Proving properties | 2025 | — | — | ||
| Cryptographic Verifiability of E2E AI Pipelines | Proving properties | 2025 | — | — | ||
| FairProof | Proving properties | 2024 | ICML 2024 | — | ||
| OATH | Proving properties | 2024 | — | — | ||
| zkAudit | Proving properties | 2024 | ICML 2024 | — | ||
| zk-OPML | Alternatives to ZK | 2026 | Journal of King Saud University CIS | — | ||
| Optimistic TEE-Rollups (OTR) | Alternatives to ZK | 2025 | — | — | ||
| TEE confidential LLM inference | Alternatives to ZK | 2025 | arXiv:2509.18886 [cs.PF] | — | ✓ | ✓ |
| opML | Alternatives to ZK | 2024 | arXiv:2401.17555 | — | ||
| opp/ai | Alternatives to ZK | 2024 | — | — | ||
| Proof of Sampling | Alternatives to ZK | 2024 | — | — | ||
| Proof of Quality | Alternatives to ZK | 2024 | — | — | ✓ | |
| NVIDIA Confidential Computing (Hopper H100) | Alternatives to ZK | 2023 | NVIDIA whitepaper WP-11459-001 v1.0 (vendor document, not peer-reviewed) | — | ✓ | ✓ |
| Lightweight Sampling Proofs of Inference | Alternatives to ZK | 2026 | IACR ePrint 2026/541 | — | ✓ | |
| ZKP-VML Survey | Surveys & reports | 2025 | arXiv:2502.18535 / Artificial Intelligence Review | — | ✓ | ✓ |
| The Definitive Guide to ZKML (2025) | Surveys & reports | 2025 | ICME Labs blog | — | ||
| Decentralized ZKML Survey | Surveys & reports | 2023 | — | — | ||
| The Cost of Intelligence | Surveys & reports | 2023 | Modulus Labs whitepaper v1.2 (eprint 2026/1063) | — | ✓ | ✓ |
primary — read in the paper survey — secondhand vendor claim, or provenance unrecorded