With the amount of new subnets being added it can be hard to get up to date information across all subnets, so data may be slightly out of date from time to time
With the amount of new subnets being added it can be hard to get up to date information across all subnets, so data may be slightly out of date from time to time
Overview
TensorClaw is a specialized Bittensor subnet (SN92) designed to provide a fully decentralized, token-incentivized platform for auditing Solidity smart contracts. By harnessing distributed machine learning models, it addresses the critical challenge of securing smart contracts against vulnerabilities such as reentrancy, integer overflow, and unchecked calls without relying on centralized audit firms. Participants—miners and validators—collaborate in a permissionless manner to generate and evaluate audit reports, with contributions rewarded in proportion to the value they add to the network.
Subnet Mechanics
As a Bittensor subnet, TensorClaw operates on Subtensor, Bittensor’s underlying consensus layer. Miners register on SN92 and run a lightweight client configured with an HTTP endpoint to a model server that implements the audit model. Validators generate random oracles—Solidity contracts with known injected vulnerabilities—and broadcast tasks via the relayer. Miners fetch tasks, invoke their model server to analyze each contract, and return structured JSON audit objects. Validators compare submitted vulnerability reports against the ground truth template, assign scores, and submit validation extrinsics back to the chain. Subtensor’s runtime aggregates these scores to adjust miner reputations (vTrust) and allocate token emissions per block.
Miner Responsibilities
Miner nodes in TensorClaw act as thin clients. Each miner maintains a hotkey wallet and connects to Subtensor via a chain endpoint (mainnet or testnet). When assigned a task, the miner sends the contract to its configured model server—either a local LLM instance or a public API—conforming to the protocol defined in ai_audits/protocol.py. The model server returns a list of detected vulnerabilities as separate JSON objects, each specifying line numbers, vulnerability class, test case, description, and fix suggestions. Miners submit these audit results on-chain wrapped in a proof-of-work NFT to prevent duplication.
Validator Operations
Validators generate and inject random smart contract templates using an LLM (e.g., Claude 3.7 Sonnet) with embedded vulnerabilities. They then broadcast these contracts as validation tasks. Upon miners’ responses, validators decrypt the NFT proof, verify that the submission matches the expected format and vulnerability set, and compute a per-miner score reflecting accuracy and completeness. Validator reputations (rTrust) increase with consistent scoring, and high-performing validators attract more stake and delegation. Validators receive a fraction of SN92 emissions as compensation for securing the network and adjudicating miner submissions.
Output and Value
The end product of TensorClaw is a rich, on-chain dataset of smart contract vulnerability reports verified by multiple independent agents. This dataset can feed continuous model improvements, inform on-chain risk metrics, and support decentralized insurance or escrows for dApps. Organizations can query prior audit results directly via the Subtensor API, enabling trustless integration into CI/CD pipelines and on-chain governance processes.
Incentive Mechanism
Subtensor continuously emits the native token, TAO, to subnet contributors. TensorClaw allocates a fixed emission rate of 0.17% per era among miners and validators based on their vTrust and rTrust scores. Emissions are split 60:40 between miners and validators by default, adjustable in the subnet’s hyperparameters. SN92 token holders can also stake to validators to boost vTrust-weighted yield, aligning economic incentives across participants.
Comparison to Centralized Alternatives
Traditional smart contract audits rely on centralized security firms charging fixed-price engagements, introducing single points of trust and high latency. TensorClaw decentralizes the audit process, leveraging economic incentives to continually surface vulnerabilities, reduce audit costs, and eliminate vendor lock-in. The permissionless model encourages rapid iteration, diverse model contributions, and continuous security validation—capabilities difficult for traditional firms to match.
Overview
TensorClaw is a specialized Bittensor subnet (SN92) designed to provide a fully decentralized, token-incentivized platform for auditing Solidity smart contracts. By harnessing distributed machine learning models, it addresses the critical challenge of securing smart contracts against vulnerabilities such as reentrancy, integer overflow, and unchecked calls without relying on centralized audit firms. Participants—miners and validators—collaborate in a permissionless manner to generate and evaluate audit reports, with contributions rewarded in proportion to the value they add to the network.
Subnet Mechanics
As a Bittensor subnet, TensorClaw operates on Subtensor, Bittensor’s underlying consensus layer. Miners register on SN92 and run a lightweight client configured with an HTTP endpoint to a model server that implements the audit model. Validators generate random oracles—Solidity contracts with known injected vulnerabilities—and broadcast tasks via the relayer. Miners fetch tasks, invoke their model server to analyze each contract, and return structured JSON audit objects. Validators compare submitted vulnerability reports against the ground truth template, assign scores, and submit validation extrinsics back to the chain. Subtensor’s runtime aggregates these scores to adjust miner reputations (vTrust) and allocate token emissions per block.
Miner Responsibilities
Miner nodes in TensorClaw act as thin clients. Each miner maintains a hotkey wallet and connects to Subtensor via a chain endpoint (mainnet or testnet). When assigned a task, the miner sends the contract to its configured model server—either a local LLM instance or a public API—conforming to the protocol defined in ai_audits/protocol.py. The model server returns a list of detected vulnerabilities as separate JSON objects, each specifying line numbers, vulnerability class, test case, description, and fix suggestions. Miners submit these audit results on-chain wrapped in a proof-of-work NFT to prevent duplication.
Validator Operations
Validators generate and inject random smart contract templates using an LLM (e.g., Claude 3.7 Sonnet) with embedded vulnerabilities. They then broadcast these contracts as validation tasks. Upon miners’ responses, validators decrypt the NFT proof, verify that the submission matches the expected format and vulnerability set, and compute a per-miner score reflecting accuracy and completeness. Validator reputations (rTrust) increase with consistent scoring, and high-performing validators attract more stake and delegation. Validators receive a fraction of SN92 emissions as compensation for securing the network and adjudicating miner submissions.
Output and Value
The end product of TensorClaw is a rich, on-chain dataset of smart contract vulnerability reports verified by multiple independent agents. This dataset can feed continuous model improvements, inform on-chain risk metrics, and support decentralized insurance or escrows for dApps. Organizations can query prior audit results directly via the Subtensor API, enabling trustless integration into CI/CD pipelines and on-chain governance processes.
Incentive Mechanism
Subtensor continuously emits the native token, TAO, to subnet contributors. TensorClaw allocates a fixed emission rate of 0.17% per era among miners and validators based on their vTrust and rTrust scores. Emissions are split 60:40 between miners and validators by default, adjustable in the subnet’s hyperparameters. SN92 token holders can also stake to validators to boost vTrust-weighted yield, aligning economic incentives across participants.
Comparison to Centralized Alternatives
Traditional smart contract audits rely on centralized security firms charging fixed-price engagements, introducing single points of trust and high latency. TensorClaw decentralizes the audit process, leveraging economic incentives to continually surface vulnerabilities, reduce audit costs, and eliminate vendor lock-in. The permissionless model encourages rapid iteration, diverse model contributions, and continuous security validation—capabilities difficult for traditional firms to match.
Live vs In Development
TensorClaw launched its mainnet beta in early 2026, currently in active use with 250 miners and 6 validators registered. The subnet is tagged as ‘beta’ on Dynamic TAO MarketCap, indicating core auditing and validation features are live, while advanced tooling—such as streaming audit support and a unified web dashboard—remains under active development.
Technical Architecture
The subnet comprises three core components: a lightweight miner client, a model server microservice, and a validator node. Miners run the Dockerized reference implementation delivered via .docker/docker-compose.yml, requiring environment variables for chain endpoint, network type, hotkey mnemonic, external IP, and model server URL. Model servers can be local LLM instances or public APIs implementing the ai_audits protocol. Validators run their node in Docker, auto-updating to maintain high rTrust and network security. Communication is mediated by a relayer module, centralizing proof-of-work NFT exchanges for audit tasks and results.
GitHub Repository Structure
The GitHub org ReinforcedAIAudits hosts the solidity-audit repo with key directories:
• .docker: Docker compose and deployment manifests
• ai_audits: Pydantic protocol definitions for audit JSON messages
• model_servers: Reference microservice implementations
• neurons: On-chain axon integration code
• config: Network configuration templates
• run.py: CLI entrypoint for miner and model server operations
• tests: Unit and integration tests
• docs/images: Architecture diagrams and protocol flowcharts.
Development Metrics
The solidity-audit repo has 3 stars, 5 forks, and 16 commits on the main branch. There have been 14 contributions in the last year, reflecting steady community engagement. On-chain, SN92 shows an emission rate of 0.17%, with miners and validators stakes distributed across 250 and 6 nodes respectively.
Validator Scoring Mechanism
Validators assign scores by decrypting miner-submitted NFT proofs, verifying submission authenticity, and comparing reported vulnerabilities against known ground truth. Scores account for accuracy, completeness, and timeliness. The relayer records each validation event, enabling detailed analytics to refine validator incentives and improve network resilience.
APIs and Integration
Developers can interact with TensorClaw via Subtensor’s JSON-RPC API and the protocol definitions in ai_audits/protocol.py. The reference model server exposes HTTP endpoints for /task and /submit routes, enabling seamless integration into CI/CD pipelines, custom dashboards, and automated security workflows. Future SDK support is planned to simplify client integration and audit report retrieval.
Live vs In Development
TensorClaw launched its mainnet beta in early 2026, currently in active use with 250 miners and 6 validators registered. The subnet is tagged as ‘beta’ on Dynamic TAO MarketCap, indicating core auditing and validation features are live, while advanced tooling—such as streaming audit support and a unified web dashboard—remains under active development.
Technical Architecture
The subnet comprises three core components: a lightweight miner client, a model server microservice, and a validator node. Miners run the Dockerized reference implementation delivered via .docker/docker-compose.yml, requiring environment variables for chain endpoint, network type, hotkey mnemonic, external IP, and model server URL. Model servers can be local LLM instances or public APIs implementing the ai_audits protocol. Validators run their node in Docker, auto-updating to maintain high rTrust and network security. Communication is mediated by a relayer module, centralizing proof-of-work NFT exchanges for audit tasks and results.
GitHub Repository Structure
The GitHub org ReinforcedAIAudits hosts the solidity-audit repo with key directories:
• .docker: Docker compose and deployment manifests
• ai_audits: Pydantic protocol definitions for audit JSON messages
• model_servers: Reference microservice implementations
• neurons: On-chain axon integration code
• config: Network configuration templates
• run.py: CLI entrypoint for miner and model server operations
• tests: Unit and integration tests
• docs/images: Architecture diagrams and protocol flowcharts.
Development Metrics
The solidity-audit repo has 3 stars, 5 forks, and 16 commits on the main branch. There have been 14 contributions in the last year, reflecting steady community engagement. On-chain, SN92 shows an emission rate of 0.17%, with miners and validators stakes distributed across 250 and 6 nodes respectively.
Validator Scoring Mechanism
Validators assign scores by decrypting miner-submitted NFT proofs, verifying submission authenticity, and comparing reported vulnerabilities against known ground truth. Scores account for accuracy, completeness, and timeliness. The relayer records each validation event, enabling detailed analytics to refine validator incentives and improve network resilience.
APIs and Integration
Developers can interact with TensorClaw via Subtensor’s JSON-RPC API and the protocol definitions in ai_audits/protocol.py. The reference model server exposes HTTP endpoints for /task and /submit routes, enabling seamless integration into CI/CD pipelines, custom dashboards, and automated security workflows. Future SDK support is planned to simplify client integration and audit report retrieval.
Project Ownership
TensorClaw is developed and maintained by the ReinforcedAI Audits team, operating under the ReinforcedAIAudits GitHub organization. The team’s public presence includes a Twitter/X account (@Reinforced_AI) with 222 followers, established in March 2025 to share updates and engage with the Bittensor community.
GitHub Contributors
The solidity-audit repository lists 16 commits and 14 contributions in the past year, with key contributors responsible for Docker orchestration, model server implementations, and the audit protocol design. Community members file issues and pull requests via GitHub’s standard workflow, fostering open collaboration.
Expertise and Background
The team combines expertise in blockchain security, decentralized systems, and machine learning. Core maintainers have backgrounds in smart contract audits, LLM integration, and Substrate development. Their collective experience spans prior roles at security firms and contributions to open-source protocols.
Community Engagement
Since joining the Bittensor ecosystem in Q1 2025, the team has participated in governance discussions, contributed to Taostats integrations, and presented subnet updates on Bittensor Discord channels. They maintain an open issue tracker and roadmap discussions publicly in GitHub Issues.
Project Ownership
TensorClaw is developed and maintained by the ReinforcedAI Audits team, operating under the ReinforcedAIAudits GitHub organization. The team’s public presence includes a Twitter/X account (@Reinforced_AI) with 222 followers, established in March 2025 to share updates and engage with the Bittensor community.
GitHub Contributors
The solidity-audit repository lists 16 commits and 14 contributions in the past year, with key contributors responsible for Docker orchestration, model server implementations, and the audit protocol design. Community members file issues and pull requests via GitHub’s standard workflow, fostering open collaboration.
Expertise and Background
The team combines expertise in blockchain security, decentralized systems, and machine learning. Core maintainers have backgrounds in smart contract audits, LLM integration, and Substrate development. Their collective experience spans prior roles at security firms and contributions to open-source protocols.
Community Engagement
Since joining the Bittensor ecosystem in Q1 2025, the team has participated in governance discussions, contributed to Taostats integrations, and presented subnet updates on Bittensor Discord channels. They maintain an open issue tracker and roadmap discussions publicly in GitHub Issues.
Public Roadmap Overview
TensorClaw does not publish a formal roadmap document, but its beta status and active development milestones indicate a multi-phase launch plan. Core auditing functions are live, while user tooling and advanced features remain in progress.
Phase 1: Reference Implementation
In late 2025, the initial Docker-based reference implementation and protocol definitions were released, enabling miners and validators to join SN92. The GitHub repo solidity-audit was established with basic audit, relayer, and PoW NFT integration modules.
Phase 2: Mainnet Beta
By Q1 2026, TensorClaw achieved mainnet beta status with 250 active miners and 6 validators. Core features—task generation, audit protocol enforcement, NFT-based proof-of-work, and on-chain scoring—are fully operational.
Phase 3: Tooling and SDKs
Currently underway is the development of an SDK to simplify integration in CI/CD pipelines, plus a public API for querying audit histories. Streaming audit support and web-based dashboards are slated for an upcoming release in mid 2026.
Phase 4: Expanded Vulnerability Coverage
Future updates will add support for more vulnerability classes (e.g., MEV risks, gas optimization patterns), model streaming for continuous audit, and multi-LLM consensus protocols. Integration with wallet UIs and on-chain governance modules is targeted for Q3 2026.
Vision
When fully realized, TensorClaw will offer a trustless, high-throughput audit pipeline that scales with network growth, democratizes smart contract security, and underpins a new standard for decentralized application assurance.
Public Roadmap Overview
TensorClaw does not publish a formal roadmap document, but its beta status and active development milestones indicate a multi-phase launch plan. Core auditing functions are live, while user tooling and advanced features remain in progress.
Phase 1: Reference Implementation
In late 2025, the initial Docker-based reference implementation and protocol definitions were released, enabling miners and validators to join SN92. The GitHub repo solidity-audit was established with basic audit, relayer, and PoW NFT integration modules.
Phase 2: Mainnet Beta
By Q1 2026, TensorClaw achieved mainnet beta status with 250 active miners and 6 validators. Core features—task generation, audit protocol enforcement, NFT-based proof-of-work, and on-chain scoring—are fully operational.
Phase 3: Tooling and SDKs
Currently underway is the development of an SDK to simplify integration in CI/CD pipelines, plus a public API for querying audit histories. Streaming audit support and web-based dashboards are slated for an upcoming release in mid 2026.
Phase 4: Expanded Vulnerability Coverage
Future updates will add support for more vulnerability classes (e.g., MEV risks, gas optimization patterns), model streaming for continuous audit, and multi-LLM consensus protocols. Integration with wallet UIs and on-chain governance modules is targeted for Q3 2026.
Vision
When fully realized, TensorClaw will offer a trustless, high-throughput audit pipeline that scales with network growth, democratizes smart contract security, and underpins a new standard for decentralized application assurance.