Evidence requirements for AI actions
Hive gives enterprises portable, verifiable records for AI usage, routing, inspection, and settlement, and we never hold your data, your funds, or your model risk.
The sample receipt is signed off-chain and verifies offline. Any chain anchor must be separately selected and specified. You keep the PDF.
You can't audit something that left no record.
Enterprises now run AI inside legal review, claims adjudication, code generation, financial reporting, vendor onboarding, and exec drafting. None of it leaves a record an external auditor can verify. That is the gap the EU AI Act, NIST AI RMF, and state attorneys general are about to close.
No record of what the AI did
Prompts, model versions, retrieval sources, and outputs disappear into vendor logs you don't own and can't subpoena. Discovery requests will arrive anyway.
No proof of who signed off
If an AI output went into a customer decision, a regulator will ask which human reviewed it and when. "Probably Steve in compliance" is not an answer that survives a deposition.
No portable evidence
Logs locked inside an LLM vendor's console are not evidence. They aren't signed, can't be exported under chain-of-custody, and aren't accepted by external auditors.
No retention policy that holds up
SOC 2, ISO 27001, HIPAA, and the EU AI Act all assume your AI use generated artifacts you can produce 6 months later. Today, almost no enterprise can.
Capture a signed record of a selected AI action.
Every AI call your enterprise makes (chat, agent, pipeline, automation, or copilot) produces a Hive Receipt: a structured, dual-signed envelope with the prompt hash, model version, output hash, retrieval sources, signer identity, and timestamp. The hash is signed off-chain and verifiable offline; any chain anchor must be separately selected and specified. The full payload stays inside your perimeter.
Your AI call runs as normal
It doesn't matter if you run OpenAI, Anthropic, Bedrock, or an internal model. Hive sits on the wire as a receipt issuer, not a proxy.
Hive issues a signed receipt
Ed25519 + ML-DSA-65 (post-quantum). ALCOA fields: Attributable, Legible, Contemporaneous, Original, Accurate. Plus signer pubkey, model/version, prompt/output hashes.
Hash, sign off-chain, verify offline
The standard receipt is signed off-chain and verifiable by your auditor without contacting Hive. Any chain commitment is optional and must be separately selected and specified; USDC payment is a separate settlement record.
You hold the artifact
PDF and JSON export, retained in your systems on your retention schedule. Hive holds the verifier, not your data.
Hive AI Governance Pilot
Production-grade receipt infrastructure for your highest-risk AI use case. 30 days, fixed scope, fixed price. We do the integration. Your team keeps the artifacts.
Prefer to scope first? The sample receipt is a live demo. It's signed, anchored, and downloadable in 30 seconds. Try it now.
30 days, 4 deliverables, audit-ready.
| Deliverable | What it is |
|---|---|
| Receipt issuance for one production AI use case | Hive instruments the AI surface you choose, whether that's legal review, claims, code, or exec drafting, and issues a signed receipt per call. Average integration: 5 business days. |
| Anchor-chain proof | Each receipt hash signed off-chain and verifiable offline; any chain anchor must be separately selected and specified via Hive's treasury. Public verifiability without exposing payload. |
| Compliance export pack | SOC 2 / ISO 27001 / HIPAA / EU AI Act crosswalk. PDFs, JSON, anchor links, control mapping notes. |
| Monthly governance report | 4-page signed PDF for your compliance committee. Volumes, routing, signers, anchors, anomalies, recommendations. |
The structural answer to "who watches the AI?"
No custody
We never hold your data, your prompts, your outputs, your model keys, or your funds. We hold the verifier and the public anchor. That's it.
No plaintext exposure
Receipts contain hashes, not content. Your IP stays inside your perimeter. Auditors verify integrity without reading payload.
Post-quantum from day one
ML-DSA-65 (NIST PQC standard) alongside Ed25519. Receipts you issue today survive the cryptographic transition the rest of your stack is dreading.
Vendor-neutral
Works across OpenAI, Anthropic, Bedrock, Vertex, internal models, and agent frameworks. Hive is the receipt layer, not the model layer.
Built for inspection
Every artifact we ship was designed to be handed to an auditor, regulator, or board member who is not technical and not friendly. It still has to hold up.
Fixed pricing, fixed scope
$10K/mo, 30 days, named deliverables. No procurement war. No SOW negotiation cycle. Start Monday, deliver by month-end.
What week-by-week looks like.
Scope & instrument
Pick the AI use case. Define routing policy. Hive provisions receipt issuance and anchor wallet. Your team installs the wire-level shim or SDK call.
First production receipts
Receipts flow. Dashboard goes live. Signers are enrolled. Standard signatures verify offline. Any selected chain-anchor profile is specified separately before deployment. Compliance starts validating against ALCOA.
Audit export dry-run
We generate the export pack as if your auditor walked in today. Your compliance lead reviews. We fix gaps. Repeat until clean.
Governance report & renewal decision
Signed monthly report delivered to your compliance committee. Renewal call: continue at $10K/mo, expand scope, or walk away with the artifacts you already own.
What your CISO will ask first.
Does Hive see our prompts or outputs?
No. Receipts contain hashes, not content. Plaintext never leaves your perimeter unless you explicitly choose to attach a portion for inspection.
What if our auditor needs to verify a receipt 3 years from now?
They re-hash the artifact and verify the off-chain signature against the published key. If a chain anchor was separately selected and specified, they can also confirm that commitment. No call to Hive is required.
Is this an MCP server or a client-side library?
Both, plus a wire-level shim. Pick the surface that fits your stack. We support agent frameworks (LangGraph, CrewAI, plain SDK calls) and direct API instrumentation.
What jurisdictions does this cover?
The receipt format is jurisdiction-neutral. We map it to EU AI Act high-risk categories, NIST AI RMF, ISO 42001, SOC 2 CC, HIPAA technical safeguards, and US state AI accountability laws as they land.
Why on-chain? Why not just signed PDFs?
Standard receipts are signed off-chain and verify offline. A separately selected and specified chain commitment can add a public checkpoint, but it does not verify the signature or make Hive institutionally independent. True independence requires non-Hive key and log custody.
What happens if Hive disappears?
You still have the PDFs, the JSON receipts, the anchor tx hashes, and the public verifier code. The proof is portable. We are deliberately not in the critical path.
See a Hive Receipt before you read the SOW.
Describe the AI use case in one sentence. We generate a signed, anchored sample receipt in under 30 seconds. It's the artifact your board will ask to see.
Standard receipts are signed off-chain and verifiable offline; any chain anchor must be separately selected and specified. USDC payment on Base is separate. We hold no data or funds.