AI verification platform EZKL.xyz builds tools to prove AI models produced outputs correctly

EZKL.xyz – Squarespace Domains customer – (United States)

Teams working across artificial intelligence and machine learning use .xyz domains to showcase platforms, tools, and services for a growing range of applications. Crypto payment platform AEON.xyz develops payment infrastructure for AI-driven transactions. Women-led hospitality platform AIVATech.xyz builds AI-powered intelligence for hotel operations. Open-source hardware company Foundation.xyz builds dedicated devices for Bitcoin, identity, and AI agent authorization. In this week’s #AIMonday, we’ll introduce you to a platform focused on verifiable AI and zero-knowledge proofs: EZKL.xyz.

Verifiable AI without revealing models or data

EZKL.xyz is the online home of EZKL, a company that builds tools designed to verify that AI models produce outputs correctly, without an outside party needing to access the model or the data it runs on. “Verifiable AI” is an approach to machine learning that prioritizes mathematical proofs and transparency over blind trust. EZKL went through web3 accelerator Beacon’s S24 cohort, which launched in September 2024.¹ As of a September 2024 press release announcing EZKL’s admission to Beacon’s accelerator cohort, the company stated it was generating tens of millions of proofs monthly across a range of delegated and decentralized systems.² EZKL.xyz describes the platform as the only audited library for verifiable AI. EZKL’s tools rely on zero-knowledge proofs, a cryptographic method that allows one party to demonstrate that a computation was carried out correctly without revealing the underlying model or data itself. As indicated in the EZKL documentation, this approach is designed to apply to three scenarios: running a public model on private data, running a private model on public data, and running a public model on public data where computation is too costly to perform directly on a blockchain.

Multiple approaches to AI verification

The platform’s documentation states that the company offers two core products: the ezkl library and Lilith, an orchestration platform for generating proofs at scale. Lilith was built to address a common constraint, according to the documentation: running larger or more complex models through EZKL can max out the computing power of a developer’s own desktop or laptop. Lilith is described as a compute cluster that developers can access through a command-line interface or REST API to generate proofs remotely, mirroring the experience of running EZKL locally.

The ezkl library converts machine learning models, exported in the ONNX format widely used across the machine learning community, into zero-knowledge circuits built on Halo2, a proving system originally developed by the Electric Coin Company, the team behind Zcash. Developers can access the library through a command-line interface, or through Python, JavaScript, and Rust bindings. The library is free to install and run locally, though the documentation notes that commercial use requires separate licensing. The website describes three ways to generate this verification, depending on a developer’s environment: pure software, designed to run on any device, including browsers and phones, without requiring changes to existing deployments; Intel TEEs (trusted execution environments), aimed at applications that require low latency; and NVIDIA CUDA, specialized kernels designed to detect GPU hardware failures and data corruption directly on-chip.

Founders with academic and machine learning backgrounds

EZKL was co-founded by Jason Morton and Alexander “Dante” Camuto. According to Beacon’s profile, Jason previously worked as a tenured mathematics and statistics professor, received a DARPA Young Faculty Award for quantum-inspired cryptographic algorithms, and served as a principal investigator in DARPA’s original Deep Learning program in 2011. Jason also previously served as Head of Cryptography at Lit Protocol, a multi-party computation-based programmable keypair platform, and earlier in his career worked in M&A investment banking.³

According to Dante’s Beacon profile, he previously founded content delivery company Myel, which was acquired by Protocol Labs, and he holds a PhD in machine learning from the University of Oxford, where he was also a visiting researcher at the Alan Turing Institute.⁴

EZKL.xyz: A pioneering domain for an AI verification platform

EZKL is building at the intersection of cryptography and machine learning, bringing together concepts and terminology from both fields. That kind of forward-looking work fits naturally with .xyz, a domain built for the next generation of innovation. The .xyz domain’s open, generic nature works across industries, giving a technical platform like EZKL room to define its own identity. EZKL.xyz gives the company a straightforward online identity that reflects its name and its focus on leading-edge technology. You can learn more by following the platform on X/Twitter @ezklxyz and LinkedIn, and by visiting EZKL.xyz.

^1. https://www.0xbeacon.com/companies/ezkl
^2. https://decrypt.co/249162/following-3-unicorns-in-2-years-web3-accelerator-beacon-launches-its-largest-cohort-to-date
^3. https://www.0xbeacon.com/companies/ezkl
^4. https://www.0xbeacon.com/companies/ezkl