Zkonduit
The company that makes EZKL, a library for proving in zero knowledge that a neural network produced a given output.
Zkonduit Inc. develops EZKL, which its documentation describes as "a developer-friendly system for verifiable AI and analytics" 1. Its verification work is EZKL:
- EZKL compiles a model exported in the ONNX format into a halo2 circuit, so a prover can show that a model produced an output while keeping either the model or the data private 2. See zero-knowledge proofs of inference.
- Two of the authors of South et al. are from EZKL. The paper uses EZKL to attest that a model with private weights reaches a stated score, with proofs for models of up to about a million parameters 3.
- Trail of Bits reviewed EZKL for Zkonduit in January 2025 and reported 34 findings, 8 of high severity. Its March fix review marked all eight high-severity findings resolved, three other findings partially resolved and two unresolved 4.
On this page
Implementations
Implementations this organization develops.
- EZKL is a library from Zkonduit that turns neural networks into zero-knowledge circuits, so a prover can show an output came from a committed model.
Publications
Sources this organization authored or published.
- BZkonduit Inc. (2026). zkonduit/ezkl (GitHub repository). GitHub. Source recordCited by EZKL; Zkonduit
- BZkonduit Inc. (2025). EZKL documentation (overview). EZKL documentation. Source recordCited by Zkonduit
- BT. South et al. (2024). Verifiable evaluations of machine learning models using zkSNARKs. arXiv. Source recordCited by Zero-knowledge proofs of inference; EZKL; Zkonduit
Sources
- BZkonduit Inc. (2025). EZKL documentation (overview). EZKL documentation. Source recordSupports: self-description of EZKL; Zkonduit copyright · documentation home
- BZkonduit Inc. (2026). zkonduit/ezkl (GitHub repository). GitHub. Source recordSupports: EZKL: ONNX models compiled to halo2 circuits; public or private model and data; Zkonduit Inc. as maintainer · README
- BT. South et al. (2024). Verifiable evaluations of machine learning models using zkSNARKs. arXiv. Source recordSupports: two EZKL authors; verifiable evaluations with EZKL; model sizes up to about a million parameters · author list; §6.1 Table 1
- BF. Casal et al. (2025). Zkonduit EZKL Security Assessment. Trail of Bits (prepared for Zkonduit Inc.). Source recordSupports: Trail of Bits audit and limited fix review: 34 findings, 8 high resolved, 3 other findings partially resolved, 2 unresolved