Core Technology

How ZKML Works

The core of Raze Protocol's innovation is

Zero-Knowledge Machine Learning (ZKML). At a high level, ZKML allows a system to prove that a piece of data was processed correctly by a specific machine learning model without revealing the private input data or the model's parameters. The process begins by compiling an AI model, such as a neural network, into a ZK-friendly

arithmetic circuit. This circuit is a series of mathematical gates that represent the model's logic. At runtime, the AI agent performs an action on its private data, and in parallel, this process generates a

Zero-Knowledge Proof. This proof is a succinct, cryptographic statement that attests that the output is the result of running the exact, approved model on the given input. The proof itself is computationally expensive to generate but is incredibly fast to verify. This transforms AI decisions from a black box into a provable mathematical fact.

The ZK Proof Flow

The ZK Proof Flow is a multi-step, trust-minimized process that guarantees the integrity of an AI agent's decision. It begins with the

Input Private Data. This could be anything from a user's credit score to a DAO's treasury data. This data is kept completely private and is never exposed on-chain. The data is then fed into the

Model Inference step, where the AI model executes its logic on the private data to produce an output. Following this, the

ZK Proof Generation component takes the output and the model's execution as inputs to create the cryptographic proof. This proof is then transmitted to the blockchain. The

Blockchain Verifier ZKML contract, which is a Solidity contract automatically generated by the ezkl framework, validates the proof's integrity and correctness on-chain. Upon successful verification, the output is accepted as a

Trustless Privacy-Preserving Decision, ensuring the action is both legitimate and private.

Why ZKML Creates Trust

ZKML creates trust by allowing anyone to verify an AI agent's decision without having to "peek under the hood". This cryptographic guarantee addresses the critical problem of opacity that plagues traditional AI.

• Mathematical Certainty: Raze Protocol replaces subjective trust with objective, mathematical certainty. The proof guarantees that "the output equals Model(input)". This means participants can be sure the AI's logic was followed exactly, preventing unauthorized actions or hidden backdoors.

• Privacy Preservation: Sensitive inputs, like personal financial or medical records, are never exposed on-chain. The proof simply confirms that the AI's decision was correct based on the data, without revealing the data itself. This brings together verifiability and privacy in a way that was previously impossible.

• Auditability: Every AI action becomes an immutable, auditable record on the blockchain. This creates a new standard for transparency and security, allowing for regulatory compliance and external audits that are backed by a cryptographic guarantee of correctness, not just a self-reported claim.

By giving users a trustless, verifiable record of an AI's actions, Raze Protocol eliminates the need to blindly trust and establishes a new standard for AI automation in the Web3 ecosystem.

Sources