Knowledge Base
Receipts · Proof-state · ProvenanceReference library
Reference library · Explainers · Field notes
The Hive Knowledge Base
A reference library for provable AI: cryptographic receipts, proof-state, and independent provenance for AI agents, machines, and payments, with the public record of our regulatory filings alongside it.
These pages explain the primitives Hive builds on and how they fit together. The foundations cover why an independent receipt is different from an internal log, what proof-state means, and how a receipt layer can witness a workflow without slowing it down. From there, the library follows the receipt across the places it applies: AI agents, machines and robots, and payments, and into the regulated settings where an independently verifiable record is the point.
Everything here is written to stand on its own and to be useful to anyone reading it, including competitors and partners. If you would like to cite, fork, or republish, please do. Attribution is appreciated but not required.
Foundations
Hub
AI Receipts: the proof layer for agents, model calls, and x402 payments
One page that maps every thread (AI receipts, agent receipts, signed model calls, x402 AI payments, MCP and API agent receipts, and the AI audit trail) to the live route that proves it.
Open the hub →
Canonical
Logs Are Claims. Receipts Are Proof.
Internal logs tell a story from inside the system. Cryptographic receipts create independently verifiable proof of what happened, what was observed, and where the proof boundary ends. That is the difference that matters for agents, machines, payments, and regulated workflows.
Read post →
Canonical
Proof-State: The Missing Primitive for AI Systems
AI systems need a machine-readable state for what has been proven, what was only claimed, and what actions remain permitted. Proof-state defines the evidence tiers (self-attested, relay-observed, chain-verified) and why honesty about the weakest boundary is the feature.
Read post →
Explainer
What Cryptographic AI Receipts Are, and Why Inference Needs Proof
A plain explanation of the primitive: signed, independently verifiable proof that an AI inference happened the way it is claimed. Why a database row is not proof, and how anyone can verify a receipt offline with ML-DSA-65.
Read post →
Architecture
AFiR Runs Beside the Control Loop, Not Inside It
Won't signing everything slow the system down? AFiR runs as a sidecar witness beside agents, robots, and payment rails, signing event boundaries asynchronously and measuring receipt latency separately from control-loop or rail latency.
Read post →
Decision layer & proof primitives
Explainer
Carnac™: The Decision Layer That Reads Before Consequences
Carnac reads a request, its output, and any proposed action before a consequence lands, scores how much the decision matters, routes it in proportion to the stakes, and signs the decision. One integration that conducts the Canon primitives underneath.
Read post →
Explainer
CarnacPrompt™: Decision at Prompt Formation Time
CarnacPrompt works before inference, inside your boundary. It classifies and structures what enters the prompt window, binds policy and context, and helps decide whether a request should proceed, all before the model is ever called.
Read post →
Explainer
S2S: Binding Inference Proof to the Silicon That Ran It
The compute-side attestation tier. Where a confidential-computing GPU is available, S2S binds inference receipts to independent, silicon-rooted attestation signed with ML-DSA-65. Honest about what is built today versus what needs live hardware to finish.
Read post →
Explainer
R3Pv: Turning Many Receipts Into One Signed Proof Vector
R3Pv groups many receipts into one signed proof vector that reports verification depth, the weakest proof boundary, policy and healing state, and the next action it permits. Proof-state scaled from a single event to a whole workflow.
Read post →
Agent receipts
Machine receipts
Payment receipts
Explainer
Stablecoin Transactions Need Receipts, Not Just Logs
A stablecoin transfer settles in seconds, but the context around it (intent, actor, policy, evidence, settlement state) lives in a log the payer controls. Receipt Relay signs that context into an independently verifiable record, with honest evidence tiers and offline verification.
Read post →
Explainer
x402 Agent Payments: Receipts for Agentic Commerce
When agents pay for AI actions on their own, the payment has to settle and the economic event has to be provable. x402 handles settlement on Base; Hive receipts make the action verifiable.
Read post →
Data & regulation
Explainer
Human-Origin Attestation: Third-Party Proof of Expert-Data Provenance
Buyer-verifiable proof that data was human-conditioned: expert-data provenance you can check without trusting the platform's own claim. Why AI data provenance needs independent third-party proof.
Read post →
Explainer
AI Assurance for Regulated Workflows: Finance, Healthcare, Legal, and Government
In regulated settings, "the model did the right thing" is not a defense. A verifiable record is. How signed inference and agent receipts map to record-keeping, oversight, and audit obligations across domains.
Read post →
Comment Letter
Re: AML & Sanctions Compliance Framework for PPSIs under the GENIUS Act
Hive Civilization's public comment on the Treasury / FinCEN / OFAC Notice of Proposed Rulemaking implementing the AML and sanctions provisions of the GENIUS Act for permitted payment stablecoin issuers. Comments close June 9, 2026.
Read letter →
Position Paper
Re: Practical Implementation of Articles 12, 13, and 14 of the EU AI Act for Agentic Commerce
A position letter to the European AI Office, the EU AI Board, and the National Competent Authorities proposing the signed transaction receipt as a candidate reference implementation for the record-keeping, transparency, and human-oversight obligations applicable to high-risk agentic-commerce systems.
Read letter →
Every post here connects back to a live primitive you can inspect: Receipt Relay for signing events at explicit evidence tiers, the Hive Ledger for searchable, offline-verifiable records, and the AI Receipts hub as the map that ties every thread to the route that proves it.