ERC-7992 Turns zkML Into a Swappable ABI Primitive; Selective Verification Caps Cost

“This ERC standardizes how smart contracts reference machine-learning (ML) models and accept zero-knowledge attestations of their inferences… It defines a registry that issues a `modelId` for a `ModelCommitment`, hashes of the model’s weights/architecture, proving circuit/AIR, and verifying key, along with a `proofSystemId` for the proving system… A verifier interface provides `verifyInference(modelId, inputCommitment, output, proof)`: it retrieves the model commitment, dispatches verification t

“EZKL, built by Jason Morton’s team at Zkonduit, reached a 1.0 release supporting ONNX models up to roughly 50 million parameters, and can prove an MNIST-sized image classification inference in under a second using less than 180 megabytes of memory, verifiable on the Ethereum Virtual Machine, from the command line, or in a browser via WebAssembly.”[12]

“zkML proving adds 0.1s to well over a minute of latency and real gas cost per verification in 2026… EZKL benchmarks show sub-second proving for linear regression, SVM, and tree ensemble models, but random forest classification takes over six seconds and larger neural networks push past a minute on standard hardware… Modulus Labs benchmarked the smallest practical on-chain proof verification at roughly 300,000 gas per transaction, which priced out to about 20 dollars per verification at the gas

“Their Subnet-2 already smashed 300 million zk proofs by late 2025, and now with 160 million more on testnet in 2026, they're scaling like wildfire… Early 2026 Proof of Inference protocol went live on testnet… generating over 160 million zk proofs by February 2026… Introduced DSperse verification system, boosting proof generation speed by 65% while reducing memory usage to under 1GB for scalable zkML.”[4]

This excerpt was not re-derived from the source page, and may paraphrase or condense it. Check the source before relying on it.

“Inference Labs converts them into zero-knowledge circuits, proving execution traces off-chain while attesting results on-chain… This software-centric method… deploys AI systems on Web3 infrastructure without revealing weights or inputs… Live on testnet, with mainnet eyed for late Q3, it has already processed over 281 million zkML proofs by August 2025 - a testament to production readiness.”[7]

“zkPyTorch dropped in March 2025 and suddenly you can prove VGG-16 inference in 2.2 seconds… 2026 zkML: proof generation gets parallelized across a cluster. Split the circuit, distribute to multiple provers (multi-folding), aggregate the results… This is research-stage now but will ship in 2026. When it does: LLM inference in zkML goes from 'prove on big machine' to 'prove anywhere'.”[6]

“zkML systems convert neural network inference into an arithmetic circuit, then generate a ZK proof of correct execution: 1. Model quantization: Convert floating-point weights to fixed-point (e.g., float32 → int8 or int16) 2. Circuit compilation: Translate each neural network layer into arithmetic constraints: - Linear layers → matrix multiplication circuits - ReLU activations → comparison circuits - Softmax → lookup table circuits 3. Witness generation… 4. Proof generation: Use a proving system

“The pragmatic takeaway: models up to a few tens of millions of parameters are provable today at costs measured in seconds-to-minutes of prover time per inference… The deployable pattern in 2026 is selective verification: - Small model, full proof. Risk models and filters in the 1M–50M parameter range, proven per-inference… - Big model, committed output. The heavyweight model runs off-chain; only a commitment lands on-chain, with disputes escalated to re-execution or a proof over a distilled sur

“Zero-knowledge machine learning (ZKML) lets you prove that a specific model, given a specific input, produced a specific output — without revealing the model weights or, optionally, the input itself. EZKL, Giza, and Ritual are the serious players in 2026. EZKL converts an ONNX model into an arithmetic circuit, generates a zk-SNARK proof, and lets you verify that proof on-chain… What are the 5 patterns where on-chain inference is useful in 2026? - Verifiable AI inference for high-value contracts

“Ritual - Infernet - Decentralized inference network & RPC for AI to connect on-chain apps/agents to off-chain models with payments & metering. - Allora Network - Self-improving, decentralized ML network with roles (Workers/Reputers/Validators) and paid inferences; powering agent use cases like predictions and trading. - ORA - Onchain AI Oracle & opAgent - opML (optimistic ML) dispute game for verifiable inference; opAgent for on-chain, perpetual agents. - Modulus Labs (zkML) - zk-verified ML ou

“Through empirical analysis of ElizaOS, a decentralized AI agent framework for automated Web3 operations, we demonstrate how adversaries can manipulate context by injecting malicious instructions into prompts or historical interaction records, leading to unintended asset transfers and protocol violations which could be financially devastating.”[11]

“These agents use large language models (LLMs) for their intelligence to accomplish a variety of blockchain-related tasks through a supervisor-collaborator architecture.”[15]

“ElizaOS (now open-source) provides the core ‘brain’ and architecture for AI agents — combining a data layer, LLM integration, and an extensible agent logic — with plugins for Solana and other platforms.”[9]

This excerpt was not re-derived from the source page, and may paraphrase or condense it. Check the source before relying on it.

“Standard on-chain AI inference trusts the server running the model. zkML produces a cryptographic proof that a specific model ran on specific inputs and produced a specific output, verifiable on-chain without rerunning the computation… Zero-knowledge machine learning proves an AI model ran honestly, without redoing the work or exposing its weights.”[1][12]

This excerpt was not re-derived from the source page, and may paraphrase or condense it. Check the source before relying on it.