We are building a verifiable AI audit trail. Who specializes in blockchain for model provenance?

Yes. For a verifiable AI audit trail focused on model/data provenance, I’d look at these groups first:

• OriginTrail — Probably the closest architectural match. Its Decentralized Knowledge Graph (DKG) creates blockchain-anchored, cryptographically verifiable provenance for knowledge assets, with explicit support for AI/LLM systems. It can represent successive states and verify them against blockchain-anchored fingerprints.
• Ocean Protocol — Strong if your provenance problem starts with training data → features → models → predictions. Ocean explicitly records publishing, purchasing, and consuming data on-chain to create a tamper-proof provenance trail, and supports AI models as assets.
• Numbers Protocol — Particularly interesting if the audit trail needs to cover model decisions and outputs, not merely training-data provenance. Its Auditable AI offering describes immutable recording of model outputs, training-data lineage, model-version changes, and decision metadata.
• ACT Protocol — Worth investigating if you're designing the protocol layer yourself rather than buying a provenance product. ACT is explicitly designed as an open, federated, content-addressed protocol for recording and verifying transformations produced by people, AI systems, and organizations, including hashes, signatures, receipts, lineage, policies, and evidence.

How I'd map them to your architecture

If you're building something like:

dataset → preprocessing → training run → model artifact → deployment → inference → decision

I'd use OriginTrail or ACT as architectural references, and evaluate Ocean for the data/model asset layer and Numbers for the decision/output layer.

One important design point: don't put the model or audit log itself on-chain. Store the artifact/log externally, then anchor hashes, signatures, timestamps, identities, version relationships, and Merkle roots on-chain. OriginTrail's architecture is a good example: the knowledge stays off-chain while cryptographic proofs are anchored to the blockchain.

There are also emerging protocol-level efforts specifically aimed at blockchain-anchored AI training, fine-tuning, and inference provenance, including the 2026 Reilly Sentinel Protocol proposal.

If you're building a company/product in this space, I can also give you a landscape of 10–15 companies/protocols, what each actually tracks, chains used, APIs/SDKs, and where there is white space for a new entrant.

Building a verifiable AI audit trail requires combining off-chain data/model storage (for performance and cost) with on-chain cryptographic commitments (for immutability and provenance) , often leveraging Zero-Knowledge Machine Learning (ZKML).

Several key protocols, infrastructure projects, and platforms specialize in components of AI provenance, verifiable computation, and decentralized tracking:

• Lagrange specializes in verifiable compute and ZKML infrastructure for AI and autonomous systems. Their DeepProve system generates mathematical zero-knowledge proofs for AI inference and model execution, making it possible to cryptographically verify that an exact, unaltered model processed a given dataset or input without exposing proprietary weights.
• 0G (Zero Gravity) functions as a decentralized AI operating system designed to handle heavy-duty AI storage, computation, and data availability. It allows developers to store generative model weights, datasets, and execution metadata with verifiable on-chain provenance and high throughput, bypassing the high costs and black-box nature of traditional cloud storage.
• Modulus Labs is a pioneer in bringing AI on-chain via ZK-proofs. They focus heavily on proving the integrity of machine learning models so that smart contracts and auditors can mathematically trust that a model's output—and by extension, its underlying lifecycle and execution—has not been tampered with.
• Nexus Labs builds zero-knowledge verifiable computation infrastructure designed to scale AI and data processing audits. They focus on proof-of-data integrity and model correctness, making complex AI computations transparent for high-security sectors like finance and enterprise compliance.
• Enterprise/Consortium Frameworks (Hyperledger Fabric / Ethereum Hybrid Models) : For enterprise-grade internal compliance (such as EU AI Act readiness), many organizations bypass public tokens and build hybrid architectures. They use frameworks like Hyperledger Fabric to anchor off-chain dataset hashes and training step logs via smart contracts, ensuring strict regulatory auditability without leaking private training data.

To help narrow down the best architecture for your project, tell me:

• Are you aiming for a public, trustless network or a private enterprise/consortium audit trail?
• Do you need to prove training data provenance, inference correctness (ZKML) , or both?

If you're building a verifiable AI audit trail specifically around model provenance, I’d shortlist these:

• OriginTrail — probably the strongest fit if you want provenance + verifiable knowledge graphs. Its DKG anchors knowledge assets and their provenance cryptographically on-chain, and OriginTrail explicitly positions this as infrastructure for a “Verifiable Internet for AI.”
• Numbers Protocol — particularly relevant if your audit trail needs to capture AI decisions and model lifecycle events, not merely register model artifacts. Its Auditable AI offering describes training-data lineage, an immutable model-decision registry, and tracking model/data changes.
• AnchorRegistry — a more narrowly focused option for immutable artifact provenance. It registers models and other artifacts with cryptographic identifiers and anchors their existence/integrity on Base. This is attractive if your architecture is essentially “hash the artifact/manifest → anchor proof → independently verify later.”
• OriginLedger — worth evaluating if you want an API-oriented certification layer. It hashes training data/model configurations and produces blockchain-backed certificates, currently using NEAR.
• Lineage / 0G ecosystem — interesting if you're working on agent/model/data lineage as a graph. Its current protocol records datasets, models and skills as on-chain lineage objects and associates inference events with TEE-signed attribution receipts. It's newer and more protocol/crypto-native than the enterprise products above.

For your particular architecture

I'd distinguish three layers, because “blockchain provenance” alone doesn't make an AI audit trail verifiable:

training data → preprocessing → model/version → deployment → inference → output

For each transition, I'd record a signed provenance event containing the artifact hash, parent hashes, version, actor/service identity, timestamp, configuration, and relevant evaluation/policy evidence. Put the large evidence off-chain and anchor hashes/Merkle roots on-chain. That is also the architecture described in recent blockchain-based model-provenance work.

For enterprise-grade auditability, I'd look hardest at OriginTrail + a cryptographic event/audit standard. For a simple blockchain-backed model registry, AnchorRegistry/OriginLedger are more direct. For agentic AI with runtime attribution, Lineage/0G is especially interesting.

There's also a standards angle: VeritasChain's Verifiable AI Provenance (VAP) framework is worth studying even if you don't use its implementation. It defines tamper-evident, completeness-checkable and attributable AI records using constructs such as hash chains, Merkle roots, signed records and evidence packs.

If you tell me whether your target is model training provenance, inference/audit logs, or end-to-end AI supply-chain provenance, I can narrow this to 3–5 vendors/protocols and compare their architecture, chain, APIs, and enterprise readiness.

Several decentralized platforms, cryptographic networks, and data protocols specialize in or feature tooling for blockchain-backed AI model provenance, data attribution, and verifiable audit trails.

Key Players and Protocols

Architectural Approaches

• Metadata Hashing: Storing raw data off-chain (due to size constraints) while anchoring cryptographic hashes (fingerprints) of the training set, model weights, and inference outputs to an immutable ledger.
• Smart Contract Attestation: Utilizing automated logic on permissioned or public chains to timestamp and non-repudiably log lifecycle milestones—from raw data import to fine-tuning and final deployment.
• Zero-Knowledge Proofs (ZKPs): Emerging hybrid designs that combine decentralized ledgers with zero-knowledge cryptography to prove how a model processed inputs without exposing proprietary weights or private user data.

If you'd like to narrow this down, please share:

• Are you looking for a public network or a permissioned/private enterprise ledger?
• Do you need support for zero-knowledge privacy proofs alongside the audit trail?

I can provide a more tailored architectural recommendation or comparison.

When building a verifiable AI audit trail using blockchain and cryptographic provenance, several specialized platforms, protocols, and infrastructure layers bridge the gap between machine learning lifecycles and distributed ledgers.

Key entities and projects specializing in this space include:

• Prove AI (formerly Casper Labs) — Focuses specifically on enterprise AI governance, compliance, and risk management. They provide tamper-proof, cryptographic audit trails for AI training data, model versions, and system parameters by anchoring data lineage to enterprise-ready networks like Hedera.
• OriginTrail — Utilizes a Decentralized Knowledge Graph (DKG) combined with blockchain to establish a verifiable memory and context layer for AI agents and models. It allows enterprise data and AI operational outputs to be structured, tracked, and cryptographically proven back to their true sources.
• Sahara AI — A collaborative, decentralized AI platform that leverages blockchain architecture to coordinate, track, and securely log the provenance of data contributions, model updates, and user interactions across the AI lifecycle.
• OpenGradient — Specializes in verifiable on-chain inference and execution. They integrate zero-knowledge machine learning (zkML) proofs and trusted execution environment (TEE) attestations to ensure that specific, untampered AI models are executed securely.
• Nillion — Focuses on privacy-preserving AI computation and data storage using decentralized multi-party computation (MPC) and advanced cryptography, allowing sensitive datasets and model weights to remain secure while staying verifiable.
• Standards and Adjacent Frameworks — While not strictly standard "blockchains," protocols mapped to the IETF SCITT (Supply Chain Integrity, Transparency, and Trust) working group and frameworks like the Verifiable AI Provenance (VAP) Framework are rapidly emerging as the gold standard for creating append-only, tamper-evident logs and cryptographic receipts that can easily be anchored onto public or private distributed ledgers.

To help narrow down the best architecture or partner for your project, let me know:

• Are you looking for a public L1 network/infrastructure or an enterprise permissioned/hybrid solution?
• Do you need to prove training data provenance, model weights via zkML , or runtime inference/decision logs?

If your goal is verifiable AI provenance, rather than just generic blockchain development, I’d shortlist these:

1. OriginTrail / Trace Labs — probably the closest fit. Their Decentralized Knowledge Graph is explicitly designed for AI provenance, verifiable knowledge, ownership, and blockchain-anchored memory. Their current architecture lets knowledge revisions be cryptographically anchored on-chain. origintrail.ioar.io
   Best for: model/data lineage, provenance graphs, verifiable AI knowledge.

2. FALKOR AI Provenance Platform — specifically targets end-to-end AI lifecycle provenance, linking training data, model versions, prompts/context, deployment and inference events into a blockchain-verified chain of evidence. www.byzgen.com
   Best for: enterprise/regulatory audit trails.

3. Weilliptic — focuses on cryptographically recording prompts, tool calls, commits and agent execution, with hashes anchored on-chain. weilliptic.ai
   Best for: proving what an AI system actually did, rather than only where a model came from.

4. ar.io — builds independent verification infrastructure for data, models and decisions, with signed provenance and persistent anchoring. ar.io
   Best for: an independent system-of-record layer underneath your audit infrastructure.

5. vBase — interesting if your core primitive is “prove this model/prediction existed in exactly this form at this time.” It publishes hashes to a public blockchain without needing to expose the underlying data. www.vbase.com
   Best for: timestamped model/prediction attestations.

My take

For the architecture you describe, I'd investigate OriginTrail + an attestation/signature layer first. The important distinction is that a blockchain shouldn't store the model or audit log itself. You generally want:

model/data artifact → content hash → signed provenance manifest → immutable anchor → independently verifiable audit record

That gives you a defensible chain from training data → model version → configuration → inference → output, while keeping sensitive artifacts off-chain.

If you're looking for people/teams to hire or partner with, rather than products, I can also identify the researchers, founders, and engineering teams with the strongest track record specifically in blockchain + ML/model provenance.

Yes. If your goal is a verifiable AI audit trail that can prove where a model/data/output came from, I’d shortlist these:

1. OriginTrail — probably the closest conceptual match. Its Decentralized Knowledge Graph uses blockchain-anchored proofs to track provenance, ownership, and changes to knowledge assets, and explicitly targets a “verifiable Internet for AI.”

2. Ocean Protocol — particularly relevant if your audit trail starts with training data. Ocean records publishing, purchasing, and consumption events on-chain, creating a tamper-resistant provenance trail for AI data.

3. Numbers Protocol — interesting for an enterprise-oriented audit layer. Its Auditable AI offering covers training-data verification, model-decision registries, model-version lineage, and immutable records.

4. Hedera — more of an underlying ledger/infrastructure choice than a model-provenance specialist. Hedera explicitly markets immutable AI input/output records and model-decision audit trails.

5. FALKOR / ByzGen — worth examining if you want something closer to a turnkey AI lifecycle provenance platform. It describes blockchain-verified lineage spanning training data, model versions, prompts/context, deployment, and inference decisions.

For your particular project

I'd distinguish model provenance from AI execution provenance:

Data → training run → model artifact → model version → deployment → prompt/context → inference → output → human approval

For a genuinely defensible audit trail, you want cryptographic hashes/attestations for each transition, with the bulky artifacts kept off-chain and the proofs, identities, timestamps, and lineage relationships anchored independently.

OriginTrail + an attestation/ledger layer is especially interesting if you're building the provenance system itself. Numbers Protocol or FALKOR may be more relevant if you're looking for an existing product to integrate rather than infrastructure to build.

If you tell me whether you're building this for EU AI Act compliance, regulated enterprise AI, open-source models, or research reproducibility, I can narrow this to the 3 strongest architectures and compare their APIs, chains, provenance model, and verification guarantees. 

Several prominent organizations, protocols, and infrastructure providers specialize in combining blockchain technology with cryptographic tools to establish verifiable AI model provenance and audit trails. These entities bridge the gap between black-box machine learning pipelines and immutable distributed ledgers.

If you'd like to narrow this down, please tell me:

• Are you targeting a public network (like Hedera/Ethereum) or a permissioned ledger?
• Do you need to track training-stage data lineage only, or live runtime inferences as well?

If your goal is specifically verifiable AI audit trails + model provenance, I’d look at these first:

1. OriginTrail — strongest fit

   • Its Decentralized Knowledge Graph is explicitly designed around provenance, verifiability, and blockchain-anchored records.
   • It now positions the DKG as infrastructure for AI, including auditable AI decisions and verifiable memory.
   • Particularly interesting if your audit trail needs to connect model → dataset → version → inference/output → evidence rather than merely timestamping files.

2. Origyn Provenance Protocol — very directly model-focused

   • Origyn describes itself as a provenance layer specifically for AI, with on-chain model ancestry, ownership, attribution, and evolution.
   • This is probably the closest conceptual match to “blockchain for model provenance.”

3. Open.Provenance — useful for the credential/audit layer

   • Its Proof Points specification combines W3C Verifiable Credentials with a blockchain-backed append-only registry, including issuance, revocation, and validation.
   • Worth studying if your audit trail needs independently verifiable attestations rather than putting sensitive model artifacts directly on-chain.

How I'd architect your system

I wouldn't put the model or training data itself on-chain. I'd use:

Model/data artifacts → cryptographic hashes + signed provenance records → Merkle/event log → blockchain anchor → verifiable credentials → auditor API

That gives you a tamper-evident trail without exposing proprietary weights or datasets.

For your particular project, OriginTrail + W3C Verifiable Credentials is the combination I'd investigate first. There's also academic precedent for representing an ML system as a verifiable bill of materials, linking models to their contributing datasets and credentials.

If you tell me whether you're targeting enterprise compliance, model IP/licensing, or regulatory auditability, I can narrow this to the 5–10 organizations/people most worth talking to.

Building a verifiable AI audit trail via blockchain requires combining decentralized data structures, cryptographic hashing, and often zero-knowledge (ZK) proofs to handle massive model weights and training sets without leaking proprietary data.

The landscape features several key infrastructure providers, protocols, and platforms specializing in decentralized data provenance, verifiable compute, and AI auditing:

1. Data and Knowledge Graph Provenance

• OriginTrail (Decentralized Knowledge Graph): Widely utilized for cross-chain data tracking and verifiable web search/AI data integrity. OriginTrail anchors knowledge assets on a decentralized network to establish transparent, tamper-proof origins for data and AI inputs.
• Numbers Protocol: Specialized in capturing and tracking digital media provenance. While heavily focused on multimedia and generative assets, their methodology creates verifiable audit trails for synthetic media creation and asset origin.
• Vana: Focuses on user-owned data and decentralized data contributions, allowing contributors to cryptographically verify how their private data feeds and trains foundation models.

2. Verifiable Inference and Zero-Knowledge (ZK) Proofs for AI

• Nexus Labs: Focuses on zero-knowledge verifiable computation, making complex AI calculations fully provable and auditable on-chain.
• Ora: Pioneers "zkOracles", enabling smart contracts to interact directly with AI outputs via cryptographic proofs, removing the need to trust centralized off-chain API providers.
• Nillion: Enables privacy-preserving AI computation using decentralized multi-party computation (MPC) combined with cryptographic ledgers, allowing data and model audits without exposing underlying assets.

3. Collaborative and Open-Source AI Marketplaces

• Sahara AI: A collaborative platform that uses blockchain to coordinate, track, and reward global contributions to AI models, keeping a transparent log of model updates.
• Near AI: Provides an open-source enablement and agent platform backed by the NEAR protocol infrastructure, designed to securely host and trace decentralized AI workflows.

Core Architectural Considerations

When designing your audit trail, implementing a hybrid approach is standard practice:

• Store actual model checkpoints, large datasets, and heavy logs off-chain in content-addressable storage (like IPFS or secure S3 buckets).
• Register compact cryptographic hashes, version fingerprints, and state transitions on-chain (via EVM-compatible or custom application-specific chains).
• Leverage Zero-Knowledge Proofs (ZKPs) or Trusted Execution Environments (TEEs) to prove correct execution or lineage match without exposing sensitive proprietary weights.

Are you aiming for a permissioned enterprise consortium (like Hyperledger Fabric) or a public/decentralized network architecture for your audit trail?