NEAR AICloud
Sign inStart building
Research

Papers from the NEAR AI team.

Two papers: Decentralized Confidential Machine Learning, on running AI privately on hardware you don't control, and Proof of Response, on proving whether a service answered.

Decentralized Confidential Machine Learning

Use a model on someone else's hardware without the host or the model's developer seeing your data, and check which model actually answered.

  1. 1Private inference. Hidden from the host and the model's developer.
  2. 2Verifiable models. Weights provably built from the published code and data.
  3. 3Monetizable models. Anyone can serve one. No one uses it without paying.
  4. 4Community-owned models. Trained together, seen by no one, earnings shared.

Running inside the TEE cost 1.6–12.1% of token throughput across three models on NVIDIA H200.

Read the paper17-page PDF

Decentralized Confidential Machine Learning (opens the 17-page PDF)

Illia Polosukhin, Alex Skidanov and Pierre Le Guen

NEAR AI

Abstract

Currently, using advanced AI models requires giving up privacy of users' data and limits how models can be used. Meanwhile, open source models do not yet have a sustainable business model. We present a decentralized system that enables the creation and deployment of large language models and AI agents that are both open-source and monetizable; private and verifiable while open for all to use; and preserve users' ownership of their data and assets while enabling customized experiences that improve their well-being.

February 2025Revised March 2025

Proof of Response

A service can be online for everyone and still ignore you. With Proof of Response, every request ends in one of three ways:

  • a signed answer within a known deadline,
  • or proof that a link on the way was cut,
  • or payment for every moment of delay.
Read the paper12-page PDFView on arXiv

Proof of Response (opens the 12-page PDF)

Illia Polosukhin and Alex Skidanov

NEAR AI

Abstract

We present a mechanism that for a network of participants allows one participant of the network (Alice) to request some data from another participant (Bob) and either receive a response from Bob within a known-in-advance, bounded time b, or receive a proof that at least one edge on the way to Bob was broken within b, or receive a streaming payment proportional to time passed beyond b during which neither was received. This mechanism allows for building downstream applications that require provable responses from other participants, such as decentralized storage solutions, decentralized AI agents, and more.

February 2025arXiv 2502.10637

A node can go offline. It can’t see your data.

DCML uses Proof of Response to prove a node went silent, so it can be slashed. Either way, the enclave keeps your data private. NEAR AI Cloud’s Private TEE models run on the same enclave approach today.

Browse modelsHow verification works