Darkwoods positions itself as a private AI lab built on the Robinhood Chain (EIP-155:4663), with its native token $DARK deployed at 0xdc45b71d7203c2e8baafc43a19bca0eb993bb54e. The official website states it enables ‘private AI on open weight models’, where ‘a wallet is the account, credits are USDG on Robinhood Chain, the models run in hardware enclaves and nothing is logged’ (darkwoods.ai). Its token is tracked on two explorers: RobinhoodChain Blockscout and Robin Etherscan. According to a sourced API snapshot, Darkwoods is categorized under ‘Robinhood Ecosystem’ and described as pioneering ‘decentralised, censorship resistant, private AI’ with ‘powerful open weight models in a fully private environment’ (source_2). Key evaluation questions remain: Is there independently verified evidence of live, functional integrations with Robinhood Chain beyond contract deployment? Does the claimed hardware-enclave execution model have public technical documentation or third-party validation? How does the project demonstrate user-facing privacy guarantees—beyond stated policy—across actual usage?
Darkwoods positions itself as a private AI lab running open-weight models in hardware enclaves, with user accounts tied to wallets and credits denominated in USDG on the Robinhood Chain. Its website (darkwoods.ai) states that “nothing is logged” and emphasizes privacy, censorship resistance, and decentralisation. The token contract is verified on Robinhood Chain (EIP-155:4663) at
0xdc45b71d7203c2e8baafc43a19bca0eb993bb54e, with both Blockscout and Etherscan links provided. A third-party API snapshot (source_2) reports circulating and total supply as ~905.8M tokens, last updated 2026-09-24.The primary thesis—that Darkwoods delivers verifiably private, enclave-based AI inference without logging—lacks evidentiary support in the supplied snapshot. No audit reports, technical whitepaper, enclave attestation mechanism, or runtime verification evidence is present. The website offers only declarative claims; the API snapshot contains no operational or security documentation. Jurisdictional claims (e.g., “censorship resistant”) are ungrounded in legal analysis or jurisdictional mapping. Counterparty exposure is high: USDG’s stability, Robinhood Chain’s governance, and enclave provider trust are all assumed but unverified. No recourse mechanisms for service failure, data leakage, or credit devaluation are disclosed.
Critical uncertainty remains: Does the claimed hardware-enclave execution exist, and is it independently verifiable? Without evidence of enclave attestation, model provenance, or runtime isolation, the core privacy promise cannot be assessed—not disproven, but unsupported. This absence triggers the rubric’s hard cap: low confidence and maximum score 6.
Overall score: 5/10 Confidence: Low