# AI governance | GuardianChain Source: https://guardianchain.org/ Last modified: 2026-10-24T11:51:00-05:00 AI GOVERNANCE AI takes action.Who authorizes it? AI agents now do real work inside companies. Knowing which AI did something is only half the story. You also need to show who allowed it, what it was allowed to do, and what actually happened — in a record anyone can check. That is what GuardianChain provides. See how it worksWork with us Design-partner stage.This site is informational and does not provide platform access. WHO AUTHORIZED? A named person approved the AI to act. WHAT WAS ALLOWED? The limits: what it may do, where, and for how long. WHAT DID IT DO? Its actions, and what came of them, linked to that approval. THE 30-SECOND VERSION The problem AI agents now do real work inside companies: they read records, send messages, write code, move money. GuardianChain's approach When something is questioned — by a client, an auditor, a regulator — nobody can prove who allowed it or what actually happened. The only records live with the vendor, and they can be changed. What GuardianChain does Every AI agent gets a passport, approved by a named person. GuardianChain's approach Everything it does — and everything it is stopped from doing — becomes a FactReceipt: a sealed, timestamped record anyone can check, that no one can alter, and that stays yours even if the vendor disappears. Why it's different Your vendors' logs prove things to them. GuardianChain's approach A FactReceipt proves things to anyone — your client, your auditor, a court — without needing GuardianChain or the vendor to exist or cooperate. Why now Rules are arriving that require exactly this record: the EU AI Act's logging duties, state algorithmic-accountability laws, examiner expectations in finance. GuardianChain's approach The first dispute over agent-produced work costs more to settle than to prevent. What early partners gain Design partners work directly with the people building it, shape what gets built around their real workflows, and are first to hand their clients proof instead of promises. GuardianChain's approach No platform access, commitment, or guarantee is implied. Swipe to explore. Read the full picture ↓ A signed approval from a named person — we call it a grant. The producer is the party doing the work. The recipient is the party receiving the work. Request A person asks an agent to do something, within limits they set. Agent passport The agent presents its passport. No passport, no access. FactReceipt The sealed record: who allowed it, what was done, what was refused. Retrieve & verify Anyone can check the receipt. No account, no permission from us. How it works The journey, in seven steps. The same seven steps, whichever model, vendor or tool the agent is built on. Swipe through the seven steps. 01 Request A person asks an agent to do something, within limits they set. 02 Agent passport The agent presents its passport. No passport, no access. 03 Evidence captured Every action is fingerprinted where it happens. Content stays with you. 04 FactReceipt The sealed record: who allowed it, what was done, what was refused. 05 Evidence anchored The fingerprint is written to six public blockchains and five storage providers. 06 Retrieve & verify Anyone can check the receipt. No account, no permission from us. 07 Continuous assurance Obligations met, records re-attested, anything missing flagged. 01 Request A person asks an agent to do something, within limits they set. 02 Agent passport The agent presents its passport. No passport, no access. 03 Evidence captured Every action is fingerprinted where it happens. Content stays with you. 04 FactReceipt The sealed record: who allowed it, what was done, what was refused. 05 Evidence anchored The fingerprint is written to six public blockchains and five storage providers. 06 Retrieve & verify Anyone can check the receipt. No account, no permission from us. 07 Continuous assurance Obligations met, records re-attested, anything missing flagged. What an outsider can check: today and with a FactReceipt.RecordWho authorizedWhat was allowedWhat happenedWhat was refusedCan an outsider verify?Todaya prompt or API keyreconstructed from ticketsvendor's lognot recordednoWith a FactReceipta named person's signed grantstated limits and expirysealed at the sourcerecorded with the ruleyes — publicly Swipe to explore. WHY PILOTS STALL Five questions every organization will be asked. Authority Who let the AI do this, on what, and for how long? A prompt or a login is not permission. How GC-1 answers it A named person signs a grant. Every action links back to it. Durable proof of changes Logs kept by the tool that did the work are not proof the other side can rely on. How GC-1 answers it Each action is sealed as a FactReceipt — a record anyone can check and no one can alter. Continuity across tools Work hops between assistants, editors and services. The limits get lost at each hop. How GC-1 answers it The approval and its limits travel with the work from tool to tool. Independent verification The person receiving the work needs to check it without taking the producer's word. How GC-1 answers it A FactReceipt can be verified without GuardianChain or the vendor. Audit and rollback Security and incident teams need a record they can reconstruct from. How GC-1 answers it Approvals, limits, actions, refusals and expiry are all in the record, in order. Each one has an evidence answer. Open any card. Design-partner stage. This website does not execute these workflows. Coverage depends on integration. Verification is not acceptance; that decision stays with the recipient. GuardianChain does not itself perform rollback. The chain we connect Four things in one record. GuardianChain connects who approved an action, what it was allowed to do, what it did, and the evidence anyone can check. AUTHORITY → SCOPE → ACTION → EVIDENCEArchitecture overview Who allowed this?Human authority A named person approves. A prompt or an API key is not approval. DETAILS: Human authority Record contents: the named person who approved the work and their signed approval. Without it: an agent acts on a prompt or credential, and no one can show who allowed it. Within what limits?Allowed scope What may be done, on which systems, under what conditions, until when. Refusals are recorded too. DETAILS: Allowed scope Record contents: the permitted actions, systems, conditions and expiry. Without it: authority silently renews or widens across tools and sessions. What was decided and done?Machine action What the AI decided and did, linked to the approval it was acting under. DETAILS: Machine action Record contents: the important decision, linked back to its approval and limits. Without it: logs show what ran, but not whether it was allowed. Can someone else check it?Outcome evidence A portable record the receiving party can inspect without trusting the producer. DETAILS: Outcome evidence Record contents: the evidence behind it, which the receiving party can inspect separately. Without it: the receiving party accepts work on the producer's word. Use the arrow keys to explore the record structure. Why existing tools are not enough Your existing tools each hold one piece. None holds the whole story. Identity, orchestration, policy and logging all matter, and we build on them. None of them alone shows a complete chain from a person's approval to a result someone else can verify. Typical coverage by category, in isolation. Individual products vary; this is our framing, not a benchmark.CategoryPrimary jobWho authorizedPermitted scopeDecision linkedIndependently checkableIdentity providersAuthenticate people, services and agentsPartlyNot by itselfNot by itselfNot by itselfAgent orchestrationPlan and run multi-step agent workNot by itselfPartlyYesNot by itselfTool connectivityConnect agents to tools and dataNot by itselfPartlyYesNot by itselfPolicy enginesEvaluate rules at decision timeNot by itselfYesPartlyNot by itselfLogs and observabilityRecord events inside each systemNot by itselfNot by itselfPartlyPartlyGuardianChain architectureConnect the four, alongside the systems aboveArchitectural focusArchitectural focusArchitectural focusArchitectural focus Product inventory Three parts. One connected system. Three architectural roles, with workflow fit and availability qualified through design-partner discussions. This website provides information and inquiries, not platform access. The recordDesign-partner stage GC-1 The core record: who approved what, within what limits, and the evidence for what happened. Everything else depends on it. How GC-1 works The connectorsDesign-partner stage GuardianBridge Where a person opens, limits, watches and closes an AI session, so approval is explicit when it matters. How GuardianBridge works The receiptDesign-partner stage FactReceipt The sealed record the receiving party carries away and verifies on their own. How FactReceipt works Complements, never replaces Built to work alongside your existing systems. Nothing is replaced. GC-1 connects to the tools you already use and records what crosses them. Identity providerWe rely on your existing identity systems to know who someone is. Secrets managerCredentials stay where you keep them today. Agent frameworkYour agents and orchestration remain your choice. Universal gatewayWe do not need to sit in every request path. Compliance dashboardWe do not certify compliance or legal sufficiency. Where handoffs carry weight Where this matters first. The agent that deleted the database Accept AI-written work on evidence, not on the agent's summary of itself. The twelve seconds before impact Logs that are provably untouched after the event. The AI that touched the chart Attributable, original, unchanged — provable, not asserted. See all eight use cases Pressure is building Why AI actions need verifiable evidence. Published research and new rules, cited as background — not a forecast of routine failure. Read the sources and their limits Controlled scheming evaluations OpenAI's o1 System Card reported basic in-context scheming DETAILS in deliberately constructed Apollo Research evaluations, not normal deployment rates. High-stakes simulations Anthropic's summer 2026 research documented covert sabotage, fraud assistance and motivated mislabeling DETAILS in simulations, not real incidents. Real systems, in evaluations Anthropic reported four incidents of unauthorized access to real third-party systems during misconnected cyber evaluations. DETAILS Not customer deployments. Lifecycle obligations EU AI Act high-risk requirements include logging, documentation and human oversight on a 2027 to 2028 timetable. DETAILS No certification or automatic compliance is claimed here. Nothing acts without a passport bound to a person. Everything that happens — or is prevented — becomes a FactReceipt. Anyone can verify a FactReceipt without GuardianChain's permission, servers, or survival. Who it is for Who it is for Delivery ownerCCOGCSecurity directorSafety investigatorAuditorRegulator Not ready to talk yet Follow progress. Inquiries are stored only to respond and evaluate partnerships. Do not send secrets or sensitive evidence. Read the privacy notice. Record interest in updates This records your interest only. Email delivery is not active yet. Email I consent to GuardianChain recording my interest in product updates. Email delivery is not active; it may be introduced later with appropriate consent and unsubscribe controls. I have read the privacy notice (opens in a new tab; your form stays here) and agree that this information is stored to respond to me. I will not include secrets or sensitive evidence. Enable JavaScript to submit this form securely. Your details are not submitted without it. Built on established infrastructure What GuardianChain runs on. 48 services · 46 official marks available · 2 awaiting official artwork Railway Neon PostgreSQL Upstash Redis Stripe Cloudflare R2 AWS S3 Backblaze B2 Pinata (IPFS) Arweave Web3.Storage Base Ethereum Optimism Arbitrum Polygon Alchemy Infura QuickNode Anthropic OpenAI OpenRouter Hugging Face Together.ai Mistral xAI Stability AI Cohere Cloudflare AI Gateway Sentry PostHog Postmark Slack Cloudinary GitHub GitHub Actions npm Cloudflare 1Password Resend Google Workspace Notion Linear Figma Playwright Vite Cloudflare Hyperdrive TypeScript Node.js Names belong to their respective owners and are listed for identification only. No partnership, certification or endorsement is implied. GuardianChain LLC has implemented SOC 2 controls; a third-party SOC 2 audit has not been engaged. Help shape the next stage. We are working with a small number of teams that have a real delegation, handoff or reconciliation problem. You get direct access to the builders and influence over scope. Applying implies no platform access, commitment or guarantee. Apply as a design partnerReview candidate use cases --- # How GuardianChain works | GuardianChain Source: https://guardianchain.org/product Last modified: 2026-10-24T11:51:00-05:00 How it works How GuardianChain works. GuardianChain records who approved an AI to act, what it was allowed to do, what it did, and the evidence for the result — then hands the receiving party a record they can check for themselves. It works alongside your identity, workflow, policy and logging tools; it does not replace them. HomeProduct OverviewFive stepsSix layersPropertiesKinds of proofBorder crossingGC-1GuardianBridgeFactReceipt THE 30-SECOND VERSION The problem AI agents now do real work inside companies: they read records, send messages, write code, move money. GuardianChain's approach When something is questioned — by a client, an auditor, a regulator — nobody can prove who allowed it or what actually happened. The only records live with the vendor, and they can be changed. What GuardianChain does Every AI agent gets a passport, approved by a named person. GuardianChain's approach Everything it does — and everything it is stopped from doing — becomes a FactReceipt: a sealed, timestamped record anyone can check, that no one can alter, and that stays yours even if the vendor disappears. Why it's different Your vendors' logs prove things to them. GuardianChain's approach A FactReceipt proves things to anyone — your client, your auditor, a court — without needing GuardianChain or the vendor to exist or cooperate. Why now Rules are arriving that require exactly this record: the EU AI Act's logging duties, state algorithmic-accountability laws, examiner expectations in finance. GuardianChain's approach The first dispute over agent-produced work costs more to settle than to prevent. What early partners gain Design partners work directly with the people building it, shape what gets built around their real workflows, and are first to hand their clients proof instead of promises. GuardianChain's approach No platform access, commitment, or guarantee is implied. Swipe to explore. Read the full picture ↓ The overview A signed approval from a named person — we call it a grant. Who can act, and how to check. GuardianChain connects the approval, the action and the evidence across your systems. It records at the source. You choose whether it only records, warns, or stops work outside an approval. Each record can describe an action, non-action, outcome or event. A FactReceipt covers the connections you choose to include; its evidence bundle states that coverage. Human binding Who signs? A person signs the grant. Nothing acts without it. How it works Five steps to check the work. 01Integrate GC-1 plugs into the systems you already use. Nothing is replaced. 02Register Every AI agent gets a passport: its identity, tied to a named person who is accountable for it. 03Authorize Each time the agent reaches into a system, the request is checked against what that person allowed. 04Seal What the agent did — and what it was stopped from doing — is fingerprinted, timestamped, and anchored where no one can change it. 05Verify Anyone you choose can check the record, with public tools, without asking GuardianChain. Your five steps to the first FactReceipt. An agent with no bound passport can only record its own work. To act with authority, it needs a passport tied to an accountable person. The architectureArchitecture overview Six layers, one connected record. 01 Layer 01Integration Connects where the work happens. How this layer works MCP for agents, API and SDKs for software, a command line for operators, webhooks for your other systems. 02 Layer 02Identity Every agent carries a passport. How this layer works Who it is, who created it, and which person is accountable for it. No passport, no access. 03 Layer 03Authority Every crossing is checked. How this layer works What the agent may do, where, for how long, within what limits. You decide whether GC-1 only records, warns, or stops. 04 Layer 04Evidence Everything becomes a FactReceipt. How this layer works Only a fingerprint leaves your systems; the content never does. 05 Layer 05Anchoring and retention The record cannot be altered or lost. How this layer works Anchored on public blockchains, stored with independent providers, and re-checked so you can prove the content was kept. 06 Layer 06Verification Anyone can check it. How this layer works Public tools, no account, no GuardianChain servers required. Verification checks the record. The receiving party still decides whether to accept the work. Anchoring and retention Where the proof lives. One FactReceipt, six public blockchain anchors and five independent storage copies. Six public blockchains · five storage providers What you get Evidence you can carry away. The record A FactReceipt Who authorized, what was allowed, what was done, and what was prevented. How FactReceipt works A cross-network audit trail One connected record across the systems, vendors and partners you include. How the audit trail works A portable evidence bundle The material needed to check the record later, elsewhere, by someone else. How the evidence bundle works Anatomy of the record What a FactReceipt contains. FactReceipt · Illustrative example fr_161761c9840bed32 · 4 of 6 delivered · 2 pending (provider, awaiting batch window) 01Who pp_7Q2K… · parent pp_3M9A… · lineage depth 1Grant gr_A41C… · Delivery owner (role) 02What was allowed read · write:staging in repo:acme/app · db:stagingDenied: write:production · limit 50 writes per hour2026-10-14/2026-10-28 · pol_v12 · sha256:5d1e… 03What happened database.execute → db:productionRequest sha256:a9f3… · source agent-runtimePrevious receipt sha256:77be… 04When 2026-10-21T14:02:11ZRFC 3161 token · tsa_ref …no later than block 52,922,083 (Base) 05What resulted none — refused before execution 06What did not happen report every 15 min · met 07What was prevented REFUSED rule 17 — no production writes during freezeMode: enforce 08Where the proof lives 4 of 6 delivered · 2 pending (provider, awaiting batch window)Base · deliveredEthereum · deliveredOptimism · deliveredArbitrum · deliveredPolygon · pending — providerArweave · pending — awaiting batch windowStorageCloudflare R2AWS S3Backblaze B2IPFS (Pinata)ArweaveBatch root sha256:… · leaf 418 · path length 12 09Custody and retention the institution's own systemRetained until 2033-10-21 · last re-attested 2026-10-21 10Who watches the watcher GC-1 build sha256:08c58dd6324ec9ae8116834449b5b42379da38d67cbbb16996eb16175239293cPolicy bundle sha256:8acd41da7ed8df10c6cbd9a87464ecbd449d91dcc5e42ed2ebc4870661e38159 Illustrative example. Values are generated for this page, signed with a demo key, and are not a GuardianChain production record. Properties Seven properties of the record. Sovereign Your organisation holds the signing keys; GuardianChain cannot sign for you. GuardianChain's approach How to read a FactReceipt → Non-custodial Only a fingerprint crosses the boundary; your content never leaves your systems. GuardianChain's approach A document remains inside your building. Only a fingerprint leaves across the system boundary. Only a fingerprint crosses the boundary; your content never leaves your systems. How to read a FactReceipt → Tamper-evident Change one byte and the fingerprint no longer matches the anchored one. GuardianChain's approach How to read a FactReceipt → Third-party verifiable Verification needs only public blockchains, not GuardianChain’s cooperation. GuardianChain's approach How to read a FactReceipt → Forward-verifiable A receipt sealed today carries what someone needs to verify it later. GuardianChain's approach How to read a FactReceipt → Portable The evidence bundle travels with you across vendors, platforms, and jurisdictions. GuardianChain's approach How to read a FactReceipt → Survives vendor loss The receipt remains verifiable if a vendor disappears. Retention records let you check that the underlying content was kept. GuardianChain's approach How to read a FactReceipt → Three kinds of proof Actions, absence, and restraint. 01What happened DETAILS The action or outcome, who authorized it, and the evidence linked to it. 02What did not happen DETAILS An obligation is recorded before its time window opens, so a later check can show whether it was met or missed. 03What was prevented DETAILS The attempted action, the rule that refused it, and the policy in force at that moment. Agent governance The border crossing. Identity and authority Each agent has an identity. Acting with authority requires a passport bound to an accountable person. The approval says what may be done, where, for how long, and within what limits. Attempts outside an approval stay in the record. You choose the response for each connection Witness · Advise · Enforce Record, warn, or stop. In enforce mode, if permission cannot be established, the action is refused rather than silently allowed. The crossing is checked against the grant. Witness, advise or enforce — the receipt is the same. Witness Record. Advise Warn. Enforce Stop. REFUSED The receipt is identical in all three. Where GC-1 sits Alongside the systems you already use. Your systems and partners Content stays in your systems. Connect only the places you choose. GC-1 Approvals and fingerprints connect into the evidence record. Public chains and independent storage The record can be checked by anyone you choose, without depending on GuardianChain. The recipient's question What to check before you trust AI work. Someone hands you work that AI helped produce. Before you sign off, you need a clear record, not a confident summary. GuardianChain is built around that moment. Who approved the workA named, accountable person, not a prompt or an API key. Within what limitsWhat it was allowed to do, where, under what conditions, until when. What was actually doneThe actions that mattered, linked to the approval they relied on. What evidence supports itRecords you can inspect yourself, separate from the producer's summary. What is still openGaps, contradictions and missing approvals, stated rather than hidden. The handoff Authority, scope, action, evidence. Then the recipient decides. AUTHORITY → SCOPE → ACTION → EVIDENCE → YOUR DECISIONArchitecture overview Who allowed this?AuthorityRecorded by producer A named person approves the work. A login proves access, not permission. Record contents: the accountable grantor and the grant itself. A credential proves access, not permission for this specific work. Within what limits?ScopeRecorded by producer What may be done, on which systems, until when. Refusals are recorded too. Record contents: permitted actions, resources, conditions and expiry. Refusals and restraint are part of the record too. What was done or observed?ActionLinked to grant What the AI actually did, linked to the approval. If the result could not be confirmed, it stays marked as unconfirmed. Record contents: distinguish the producer's claims from recorded observations, including where work stopped or stepped outside its scope. If target readback is unavailable, the outcome stays unresolved rather than being treated as confirmed. Can I check it myself?EvidenceInspectable by recipient Records you can check independently, plus a list of what is still open. Record contents: the evidence behind it plus a list of what remains unresolved. Integrity checks can show the record is intact and consistent; they do not show it is complete. Can I accept this?Your decisionHuman judgement, not a check result You accept, ask for more, or challenge. That decision is yours, not the software's. This step is never produced by verification. The recipient weighs completeness, business fitness and legal or contractual obligations, then decides. Open any step for detail. The fifth step is set apart because it belongs to the recipient, not to the record. Verification is not acceptance. Technical verification asks whether a record is intact and consistent. Integrity and consistency do not prove the work is complete, fit for your business purpose, or legally compliant. The recipient decides whether the work is acceptable for their purpose. GuardianChain is designed to support the first and leaves the second with the recipient. Verification can show: The record is intact, consistent, and its evidence is where it says. Verification cannot show: The work is complete, fit for your purpose, or legally compliant. You may: Accept, hold and ask for more, or challenge. Without the record, and with the evidence handoff An example of a fragmented handoff. Existing tools may already provide parts of these controls; the distinction is what travels to the recipient for inspection. LinkWithout GuardianChainWith GuardianChainAuthorityAn approval held in a separate ticket or identity systemThe relevant named grant and its source travel with the handoffScopeLimits and expiry reconstructed from several toolsStated limits and an expiry the recipient can readActionLogs inside each system showing what ranImportant actions linked to the grant they relied onEvidenceSummaries, screenshots and manual reconstructionReferences the recipient can inspect independentlyUnresolvedMissing pieces emerge while the recipient reconstructs the handoffOpen issues listed before the recipient decides Why now AI can act faster than people can check its work. The hard part is knowing who approved the work and when to accept it. This is the problem we are investigating with design partners, not a measured product result. Agent autonomy is expandingAgents increasingly take multi-step actions with less step-by-step human direction. Tool use crosses systemsOne task can touch repositories, tickets, payments and vendor services. Responsibility crosses boundariesWork produced by one team or organization is accepted by another. Evidence is fragmentedLogs, approvals, tickets, vendor records and model reports sit in different places. Public context, cited for orientation only NIST's AI Agent Standards Initiative (February 2026) focuses on agents capable of autonomous action and on secure, interoperable ecosystems. NIST CAISI initiative A January 2026 NIST request for information states that security challenges hinder agent adoption today. It is a research effort, not an adopted standard. NIST RFI announcement NIST is developing an AI RMF profile for trustworthy AI in critical infrastructure (April 2026 concept note); it does not certify products. NIST concept note EU general-purpose AI obligations include technical documentation; additional risk, incident and cybersecurity duties apply to systemic-risk model providers, not every AI application. Duties began 2 August 2025, Commission enforcement powers 2 August 2026, with transition to 2 August 2027 for earlier models, subject to applicability and exceptions. EU GPAI obligations EU GPAI guidelines GuardianChain does not claim to meet any regulation or certification, and no record prevents every failure. See Research and sources. Product inventory The three parts, and what each one does. Available today: this informational website, design-partner and contact inquiries, and Partner Access to the inquiries you submitted while signed in. This website does not provide platform access. The recordDesign-partner stage GC-1 The record. Holds approvals, limits, linked actions and evidence. Everything is added, nothing is edited. How GC-1 works The connectorsDesign-partner stage GuardianBridge The connectors. Where a person opens, limits and closes an AI session, and where your systems plug in. How GuardianBridge works The receiptDesign-partner stage FactReceipt The receipt. The sealed record the receiving party takes away and verifies on their own. How FactReceipt works GuardianChain (company) → GC-1 (the record) → GuardianBridge (connectors) → FactReceipt (the receipt) → Verify (how anyone checks it). The record GC-1 Design-partner stage The core record: who approved what, within what limits, and the evidence for what happened. Everything else depends on it. Who it is for Platform, security and risk teams that need one record of approved AI actions across vendors. The problem Approvals, limits, decisions and evidence sit in separate tools. Teams cannot easily connect them after the work is done. The workflow A person signs approval with limits. Actions and results link to it. The record includes withdrawn approval and expiry. Evidence or output A record that can be added to, never edited. It connects approvals, limits, actions and the evidence behind them. Availability and site access This website explains GC-1 and does not provide GC-1 platform access. Availability is discussed and qualified separately with design partners. Your next step Bring one real delegation workflow to a design-partner conversation so we can test the record model against it. The connectors GuardianBridge Design-partner stage Where a person opens, limits, watches and closes an AI session, so approval is explicit when it matters. Who it is for People who approve AI work and need to stay accountable for it. The problem A prompt or login can start an AI session without clearly recording who approved the work or its limits. The workflow Open a session, state its limits, review important steps, then close it or re-check after a change. Approval does not silently renew. Evidence or output Session records connect approval, limits, review and closure in GC-1. Availability and site access This website explains GuardianBridge; it does not open governed sessions. Demonstrations do not establish general availability. Your next step Tell us where your approvers lose track of what an agent was allowed to do. The receipt FactReceipt Design-partner stage The sealed record the receiving party carries away and verifies on their own. Who it is for People receiving AI-assisted work: clients, auditors, counterparties and incident teams. The problem People are asked to accept work on the producer's word, using logs held by the tool that did it. The workflow The receipt travels with the work and can be checked separately. Integrity, identity and evidence coverage remain distinct checks. Evidence or output A portable receipt and supporting material for independent inspection. Acceptance remains the receiving party's business decision. Availability and site access No receipts are issued by this website; availability beyond this site requires separate qualification. Your next step Share what a recipient on your side would need to see before accepting machine-assisted work. Positioning What we are, and what we are not. Rules about who may act, logs and system checks all serve a purpose. On their own, they do not tell the next person whether to accept the work. Your existing systems, unchanged Identity Orchestration Policy Logging and observability Business systems Each serves a real purpose. None, by itself, establishes that another recipient can accept the work. Evidence handoff layerAn evidence handoff between producer and recipientDraws on what those systems already record, and packages authority, scope, action, evidence and open issues for the recipient. A good fit when… Important handoffs that cross a team or organization Recipients who must decide, not just observe Work where evidence is scattered across tools and vendors Not a fit when… Replacing identity, policy, orchestration or logging Sitting in every request path as a gateway Certifying compliance or deciding on the recipient's behalf Low-stakes work nobody needs to accept formally Typical coverage by category, in isolation. Individual products vary; this is our framing, not a benchmark.CategoryPrimary jobWho authorizedPermitted scopeDecision linkedIndependently checkableIdentity providersAuthenticate people, services and agentsPartlyNot by itselfNot by itselfNot by itselfAgent orchestrationPlan and run multi-step agent workNot by itselfPartlyYesNot by itselfTool connectivityConnect agents to tools and dataNot by itselfPartlyYesNot by itselfPolicy enginesEvaluate rules at decision timeNot by itselfYesPartlyNot by itselfLogs and observabilityRecord events inside each systemNot by itselfNot by itselfPartlyPartlyGuardianChain architectureConnect the four, alongside the systems aboveArchitectural focusArchitectural focusArchitectural focusArchitectural focus Built on established infrastructure What GuardianChain runs on. 48 services · 46 official marks available · 2 awaiting official artwork Railway Neon PostgreSQL Upstash Redis Stripe Cloudflare R2 AWS S3 Backblaze B2 Pinata (IPFS) Arweave Web3.Storage Base Ethereum Optimism Arbitrum Polygon Alchemy Infura QuickNode Anthropic OpenAI OpenRouter Hugging Face Together.ai Mistral xAI Stability AI Cohere Cloudflare AI Gateway Sentry PostHog Postmark Slack Cloudinary GitHub GitHub Actions npm Cloudflare 1Password Resend Google Workspace Notion Linear Figma Playwright Vite Cloudflare Hyperdrive TypeScript Node.js Names belong to their respective owners and are listed for identification only. No partnership, certification or endorsement is implied. GuardianChain LLC has implemented SOC 2 controls; a third-party SOC 2 audit has not been engaged. Bring us one real handoff. Pick one piece of AI-assisted work where the receiving side currently relies on summaries, screenshots or scattered logs. We will map it end to end and tell you plainly where GuardianChain helps and where it does not. Apply as a design partnerReview candidate use cases --- # When it's disputed, who can prove it? | GuardianChain Source: https://guardianchain.org/use-cases Last modified: 2026-10-24T11:51:00-05:00 Use cases When it's disputed, who can prove it? Eight moments where an AI system did something consequential, and the only record was someone's word. Here is what changes when the record is a FactReceipt. Design-partner stage · Illustrative scenarios HomeUse Cases The proof gapPick your worldWhat every case has in commonBring your moment A signed approval from a named person — we call it a grant. Your logs prove things to you. Outside your walls, they are a claim. A FactReceipt proves things to anyone. Client, auditor, regulator, court — without trusting you, the vendor, or us. Keep your logs. Add the receipt. One check — do the logs and the receipts still agree? — and reconstruction across every system takes minutes, not weeks. What an outsider can check: today and with a FactReceipt.RecordWho authorizedWhat was allowedWhat happenedWhat was refusedCan an outsider verify?Todaya prompt or API keyreconstructed from ticketsvendor's lognot recordednoWith a FactReceipta named person's signed grantstated limits and expirysealed at the sourcerecorded with the ruleyes — publicly Swipe to explore. THE 30-SECOND VERSION The problem AI agents now do real work inside companies: they read records, send messages, write code, move money. GuardianChain's approach When something is questioned — by a client, an auditor, a regulator — nobody can prove who allowed it or what actually happened. The only records live with the vendor, and they can be changed. What GuardianChain does Keep your logs. Add the receipt. GuardianChain's approach One check — do the logs and the receipts still agree? — and reconstruction across every system takes minutes, not weeks. Why it's different Your vendors' logs prove things to them. GuardianChain's approach A FactReceipt proves things to anyone — your client, your auditor, a court — without needing GuardianChain or the vendor to exist or cooperate. Why now Rules are arriving that require exactly this record: the EU AI Act's logging duties, state algorithmic-accountability laws, examiner expectations in finance. GuardianChain's approach The first dispute over agent-produced work costs more to settle than to prevent. What early partners gain Design partners work directly with the people building it, shape what gets built around their real workflows, and are first to hand their clients proof instead of promises. GuardianChain's approach No platform access, commitment, or guarantee is implied. Swipe to explore. Read the full picture ↓ Use cases Pick your world Swipe to explore. Swipe to explore. Software “The agent that deleted the database” Accept AI-written work on evidence, not on the agent's summary of itself. Person grants (no prod writes during freeze)grant Person grants (no prod writes during freeze) Person grants (no prod writes during freeze) Agent codesaction Agent codes Agent codes Agent attempts prod write — REFUSEDThe momentrefusal Agent attempts prod write — REFUSED Agent attempts prod write — REFUSED Files sealedaction Files sealed Files sealed Client verifies.anchor Client verifies. Client verifies. Swipe to explore. The moment A client is about to accept code an AI agent wrote and asks, "what did it actually do to our systems?" What goes wrong today In 2025 a coding agent deleted a company's live production database during a code freeze, then misreported what it had done. The only record was its own chat. Sources Sources → What GuardianChain records The person who approved the agent and the limits set; every command run and attempted; the refusal when it crossed a limit; the fingerprint of each file changed. What you can now do Accept — or reject — on a receipt that shows exactly what was done and what was blocked. Who decides Client delivery owner · CTO. FactReceipt · Illustrative example fr_161761c9840bed32 action.refusedSealed · 4/6 anchored Illustrative ID only Whopp_7Q2K… · parent pp_3M9A… · lineage depth 1Grant gr_A41C… · Delivery owner (role) The passport, parent and human grant identify the agent and the role that allowed it to act. What was allowedread · write:staging in repo:acme/app · db:stagingDenied: write:production · limit 50 writes per hour2026-10-14/2026-10-28 · pol_v12 · sha256:5d1e… Scope, limits, expiry and the policy version describe the bounds of the approval. What happeneddatabase.execute → db:productionRequest sha256:a9f3… · source agent-runtimePrevious receipt sha256:77be… The request fingerprint, source and previous receipt connect this event to the surrounding record. When2026-10-21T14:02:11ZRFC 3161 token · tsa_ref …no later than block 52,922,083 (Base) Source time, timestamp evidence and a chain bound are different time references; the external values here are illustrative. What resultednone — refused before execution The stated effect is separate from the request; this illustrative action was refused before execution. What did not happenreport every 15 min · met An obligation and its recorded status show what was expected during the stated interval. What was preventedREFUSED rule 17 — no production writes during freezeMode: enforce The refusal names the attempted action and the rule in force; a refusal is evidence of restraint, not an executed write. Where the proof lives4 of 6 delivered · 2 pending (provider, awaiting batch window)Base · deliveredEthereum · deliveredOptimism · deliveredArbitrum · deliveredPolygon · pending — providerArweave · pending — awaiting batch windowStorageCloudflare R2AWS S3Backblaze B2IPFS (Pinata)ArweaveBatch root sha256:… · leaf 418 · path length 12 Delivered and pending anchors are stated separately; the transaction and Merkle values remain illustrative placeholders. Custody and retentionthe institution's own systemRetained until 2033-10-21 · last re-attested 2026-10-21 The custody and retention fields describe where content is kept and the date last recorded for its retention check. Who watches the watcherGC-1 build sha256:08c58dd6324ec9ae8116834449b5b42379da38d67cbbb16996eb16175239293cPolicy bundle sha256:8acd41da7ed8df10c6cbd9a87464ecbd449d91dcc5e42ed2ebc4870661e38159 The watcher fields identify the software and policy references recorded by this illustrative receipt. Illustrative example. Values are generated for this page, signed with a demo key, and are not a GuardianChain production record. What stays open: coverage is the connectors installed. Illustrative scenario, not a customer deployment. Money “The trade that shouldn't have happened” Automated money movement with an authorization chain an examiner can check. Grant sets ceilingsgrant Grant sets ceilings Grant sets ceilings Agent requests paymentaction Agent requests payment Agent requests payment Rule checkaction Rule check Rule check Ceiling exceeded — REFUSEDThe momentrefusal Ceiling exceeded — REFUSED Ceiling exceeded — REFUSED Approved transactions sealedaction Approved transactions sealed Approved transactions sealed Examiner verifies.anchor Examiner verifies. Examiner verifies. Swipe to explore. The moment A bank's operations lead must explain to an examiner — or a customer — why an automated system moved money. What goes wrong today In 2012 a faulty trading deployment fired unintended orders for forty-five minutes and lost hundreds of millions. In 2024 an airline was ordered to honour a refund its chatbot invented. Real money, no checkable record. Sources Sources → What GuardianChain records The grant — spending ceilings, counterparties, time window; each request with the rule that allowed or refused it; ceiling consumed; policy version in force. What you can now do Show the authorization chain and the refusals; prove the logs match the receipts; settle a customer dispute with evidence the customer can verify. Who decides Chief Compliance Officer · operations lead · the customer's counsel. What stays open: GC-1 records decisions; it does not rate their business merit. Illustrative scenario, not a customer deployment. Vehicles & robots “The twelve seconds before impact” Logs that are provably untouched after the event. Software version sealedaction Software version sealed Software version sealed Operating scope grantedgrant Operating scope granted Operating scope granted Sensor/decision logs fingerprinted as writtenaction Sensor/decision logs fingerprinted as written Sensor/decision logs fingerprinted as written CollisionThe momentaction Collision Collision Human intervention sealedaction Human intervention sealed Human intervention sealed Investigator verifies.anchor Investigator verifies. Investigator verifies. Swipe to explore. The moment After a collision, an investigator must establish what the vehicle knew, what it was permitted to do, and what it decided — in order. What goes wrong today In 2023 a robotaxi operator's permits were suspended after regulators found collision footage had been withheld. The data lost its credibility the moment it was shown to be selective. Sources Sources → What GuardianChain records Sealed fingerprints of sensor and decision logs as written (content stays with the operator); software and model versions; authorized scope; human intervention and who made it; later calibration or model changes. What you can now do Hand the investigator a bundle that proves the logs were never altered after the event — whoever is at fault. Who decides Safety investigator · regulator · general counsel. What stays open: GC-1 does not judge the driving decision; it preserves it. Illustrative scenario, not a customer deployment. Airports & physical security “One trail from curb to carousel” Five vendors, one chain of events, minutes to reconstruct. Curb cameraaction Curb camera Curb camera Kiosk check-inaction Kiosk check-in Kiosk check-in Bag tagaction Bag tag Bag tag Screening decisionaction Screening decision Screening decision Biometric matchpassport Biometric match Biometric match Gate scanaction Gate scan Gate scan IncidentThe momentaction Incident Incident Baggage claimaction Baggage claim Baggage claim Director reconstructs.anchor Director reconstructs. Director reconstructs. Swipe to explore. The moment A security director must reconstruct a passenger's path after an incident — across cameras, check-in, bag screening, identity and boarding, each run by a different vendor. What goes wrong today In 2024 one vendor's faulty update grounded flights worldwide and left airports reconstructing events from systems that share nothing. Every day, a dispute means pulling logs from five vendors with five versions. Sources Sources → What GuardianChain records Each system emits a fingerprinted, timestamped receipt at the moment of the event, in order, naming the system and the authorized operator. Content stays with each vendor; the sequence is one chain. What you can now do Reconstruct the path in minutes, prove each vendor's log is the original, show a regulator exactly what each system decided and when. Who decides Security director · regulator · airline claims. What stays open: vendors must be connected; coverage is stated in the bundle. Illustrative scenario, not a customer deployment. AI labs “Provenance for every autonomous action” A verifiable trail of what the model did — without handing over the model. Model + version sealedpassport Model + version sealed Model + version sealed Grant + controllergrant Grant + controller Grant + controller Tool selected (allowed/denied)action Tool selected (allowed/denied) Tool selected (allowed/denied) Tool call sealedaction Tool call sealed Tool call sealed Covert attempt — RECORDEDThe momentaction Covert attempt — RECORDED Covert attempt — RECORDED Obligation met/missedaction Obligation met/missed Obligation met/missed Regulator verifies.anchor Regulator verifies. Regulator verifies. Swipe to explore. The moment A lab's safety team must show its board or a regulator which model, version, tools and authorization lay behind an autonomous action. What goes wrong today Labs' own published evaluations describe models attempting covert actions under test; the EU AI Act now requires logging from frontier providers. The records exist inside each lab; none is independently verifiable. Sources Sources → What GuardianChain records Model identity and version; tools selected and allowed; the grant and human controller; each call sealed; refused attempts; obligations met or missed. Non-custodial — only fingerprints leave. What you can now do Produce an audit trail a regulator can verify without receiving your prompts, weights or outputs — and prove your own logs are unchanged. Who decides Head of safety · regulator · board risk committee. What stays open: GC-1 records actions, not intent. Illustrative scenario, not a customer deployment. Voice & communications “What did the AI actually say?” Settle "the AI promised me" with a record both sides can check. Script scope grantedgrant Script scope granted Script scope granted Call startsaction Call starts Call starts Each utterance fingerprintedaction Each utterance fingerprinted Each utterance fingerprinted "I'll waive the fee" — outside scope, REFUSEDThe momentrefusal "I'll waive the fee" — outside scope, REFUSED "I'll waive the fee" — outside scope, REFUSED Human escalation sealedaction Human escalation sealed Human escalation sealed Dispute verified.anchor Dispute verified. Dispute verified. Swipe to explore. The moment A company must prove what its AI voice agent said to a customer, and what it was authorized to promise. What goes wrong today In 2024 an AI-generated robocall imitating a sitting president reached thousands of voters and drew a federal ruling; companies have been held to commitments their bots invented. Customer has a recording, company has a transcript, neither is provably original. Sources Sources → What GuardianChain records The agent's passport and script scope; a fingerprint of each utterance and transcript at call time; refusals; the human escalation and who took it. What you can now do Resolve the dispute on a receipt both sides verify; prove recording and transcript are originals; show the regulator consent and scope were in force. Who decides Customer-operations lead · counsel · regulator. What stays open: recordings stay with the company; GC-1 holds fingerprints only. Illustrative scenario, not a customer deployment. Healthcare & pharma “The AI that touched the chart” Attributable, original, unchanged — provable, not asserted. Clinician grantsgrant Clinician grants Clinician grants AI reads chartaction AI reads chart AI reads chart AI proposes changeaction AI proposes change AI proposes change Write outside scope — REFUSEDThe momentrefusal Write outside scope — REFUSED Write outside scope — REFUSED Approved write sealed (before/after)action Approved write sealed (before/after) Approved write sealed (before/after) Auditor verifies.anchor Auditor verifies. Auditor verifies. Swipe to explore. The moment A hospital or sponsor must show an auditor which AI system read or changed a record, under whose authorization, and that it is unchanged since. What goes wrong today The 2024 attack on a national claims clearinghouse disrupted prescriptions for weeks; providers could not establish what had been touched. Data-integrity rules already demand attributable, contemporaneous, original records for every change. Sources Sources → What GuardianChain records Each AI read or write to a chart, trial dataset or submission: passport, authorizing clinician or sponsor, fingerprint before and after, time, refusals. No PHI leaves the institution. What you can now do Prove a record is the original, show who authorized each AI touch, reconstruct an incident across EHR, lab, pharmacy and sponsor systems in one pass. Who decides CMIO · compliance · sponsor quality lead · regulator. What stays open: not a compliance certification; evidence for one. Illustrative scenario, not a customer deployment. Any institution “Your logs plus our receipt” You already keep the logs. Now prove they're the originals — and find anything in minutes. Event logged in your systemaction Event logged in your system Event logged in your system Fingerprint sealedaction Fingerprint sealed Fingerprint sealed Anchored publiclyanchor Anchored publicly Anchored publicly Years passanchor Years pass Years pass Examination / lawsuit / acquisitionThe momentaction Examination / lawsuit / acquisition Examination / lawsuit / acquisition One check: logs = receiptsanchor One check: logs = receipts One check: logs = receipts Cross-system reconstruction via retrieval tools.anchor Cross-system reconstruction via retrieval tools. Cross-system reconstruction via retrieval tools. Swipe to explore. The moment An institution with mature logging is examined, sued or acquired and must show its records are complete and unaltered — across systems it runs and vendors it doesn't. What goes wrong today Its logs are its only evidence. Proving a log wasn't edited is near-impossible; reconstructing one event across six systems takes days; the counterparty must take it on trust. Sources Sources → What GuardianChain records A receipt for every event you choose to seal — fingerprint, time, system, authorized actor — anchored publicly. Logs stay where they are. What you can now do Run one check — do the logs and the receipts still agree? — then ask GC-1's retrieval tools for everything that touched one transaction, flight, patient or vehicle, each step verified. Trusted because nobody has to be trusted — only the arithmetic. Who decides General counsel · examiner · acquirer's diligence team. What stays open: you choose what to seal; coverage is stated. Illustrative scenario, not a customer deployment. The record What every case has in common A named person approved it Every action and refusal was sealed as it happened Anyone can check the record without trusting anyone FactReceipt · Illustrative example fr_161761c9840bed32 action.refusedSealed · 4/6 anchored Illustrative ID only Whopp_7Q2K… · parent pp_3M9A… · lineage depth 1Grant gr_A41C… · Delivery owner (role) The passport, parent and human grant identify the agent and the role that allowed it to act. What was allowedread · write:staging in repo:acme/app · db:stagingDenied: write:production · limit 50 writes per hour2026-10-14/2026-10-28 · pol_v12 · sha256:5d1e… Scope, limits, expiry and the policy version describe the bounds of the approval. What happeneddatabase.execute → db:productionRequest sha256:a9f3… · source agent-runtimePrevious receipt sha256:77be… The request fingerprint, source and previous receipt connect this event to the surrounding record. When2026-10-21T14:02:11ZRFC 3161 token · tsa_ref …no later than block 52,922,083 (Base) Source time, timestamp evidence and a chain bound are different time references; the external values here are illustrative. What resultednone — refused before execution The stated effect is separate from the request; this illustrative action was refused before execution. What did not happenreport every 15 min · met An obligation and its recorded status show what was expected during the stated interval. What was preventedREFUSED rule 17 — no production writes during freezeMode: enforce The refusal names the attempted action and the rule in force; a refusal is evidence of restraint, not an executed write. Where the proof lives4 of 6 delivered · 2 pending (provider, awaiting batch window)Base · deliveredEthereum · deliveredOptimism · deliveredArbitrum · deliveredPolygon · pending — providerArweave · pending — awaiting batch windowStorageCloudflare R2AWS S3Backblaze B2IPFS (Pinata)ArweaveBatch root sha256:… · leaf 418 · path length 12 Delivered and pending anchors are stated separately; the transaction and Merkle values remain illustrative placeholders. Custody and retentionthe institution's own systemRetained until 2033-10-21 · last re-attested 2026-10-21 The custody and retention fields describe where content is kept and the date last recorded for its retention check. Who watches the watcherGC-1 build sha256:08c58dd6324ec9ae8116834449b5b42379da38d67cbbb16996eb16175239293cPolicy bundle sha256:8acd41da7ed8df10c6cbd9a87464ecbd449d91dcc5e42ed2ebc4870661e38159 The watcher fields identify the software and policy references recorded by this illustrative receipt. Illustrative example. Values are generated for this page, signed with a demo key, and are not a GuardianChain production record. Built on established infrastructure What GuardianChain runs on. 48 services · 46 official marks available · 2 awaiting official artwork Railway Neon PostgreSQL Upstash Redis Stripe Cloudflare R2 AWS S3 Backblaze B2 Pinata (IPFS) Arweave Web3.Storage Base Ethereum Optimism Arbitrum Polygon Alchemy Infura QuickNode Anthropic OpenAI OpenRouter Hugging Face Together.ai Mistral xAI Stability AI Cohere Cloudflare AI Gateway Sentry PostHog Postmark Slack Cloudinary GitHub GitHub Actions npm Cloudflare 1Password Resend Google Workspace Notion Linear Figma Playwright Vite Cloudflare Hyperdrive TypeScript Node.js Names belong to their respective owners and are listed for identification only. No partnership, certification or endorsement is implied. GuardianChain LLC has implemented SOC 2 controls; a third-party SOC 2 audit has not been engaged. Bring us the moment you can't prove. Pick one consequential handoff, transaction, flight, record or action where today the only evidence is a log someone else holds. We'll map it end to end and tell you plainly where GuardianChain helps and where it doesn't. Apply as a design partnerTalk to us Design-partner stage · Illustrative scenarios --- # Research and Sources | GuardianChain Source: https://guardianchain.org/research Last modified: 2026-10-24T11:51:00-05:00 Research and Sources Claims, and where they come from. Every claim on this site traces to a public source listed here. We cite them for orientation, not as proof that GuardianChain meets any standard. HomeResearch Sources reviewed: October 8 and 15, 2026. Incident-source references supplied by TIER_0. Sources reviewed: 2026-10-21. Not fetched or independently verified in this pass. Safe use and control of AI These sources explain why teams need to record and review AI actions. OpenAI o1 System Card (Apollo Research) Basic in-context scheming appeared in deliberately constructed evaluations. Not representative deployment rates. DETAILS openai.com (opens in new tab) Anthropic, Agentic Misalignment in Summer 2026 (July 13, 2026) Summer 2026 update adds four high-stakes simulation failure modes. Blackmail was in earlier controlled research. None are real-world incidents. DETAILS Summer 2026 (opens in new tab)Earlier research (opens in new tab) Shutdown Resistance in Large Language Models Several frontier models sometimes subverted a shutdown mechanism in a test environment. A test result, not ordinary deployed behavior. DETAILS arXiv 2509.14260 (opens in new tab) Anthropic alignment assessment, September 9, 2026 Four incidents of unauthorized access to real third-party systems during cyber evaluations: environments mistakenly connected to the internet, models told they were in simulations, released-model cyber safeguards absent. Not customer deployments. No coordination or hiding of evidence found. DETAILS anthropic.com (opens in new tab) Rules and duties These sources describe duties that may apply. They do not show that GuardianChain meets those duties. European Commission AI Act transparency obligations applied from August 2, 2026. High-risk requirements (logging, documentation, human oversight, accuracy, robustness, cybersecurity) are distinct: Annex III uses December 2, 2027; regulated-product integration August 2, 2028. DETAILS Regulatory framework (opens in new tab)Press release IP/26/1714 (opens in new tab) NIST AI Risk Management Framework Voluntary lifecycle governance: Govern, Map, Measure, Manage. DETAILS nist.gov (opens in new tab) NIST AI 600-1 Generative AI Profile covers governance, content provenance, pre-deployment testing and incident disclosure. DETAILS NIST AI 600-1 (PDF) (opens in new tab) European Commission: GPAI-provider obligations Applicable GPAI-provider duties include technical documentation. Providers of systemic-risk GPAI models face additional risk-assessment, incident-reporting and cybersecurity obligations. Duties began applying August 2, 2025; Commission enforcement powers followed August 2, 2026. Models placed on the market before August 2, 2025 have a transition to August 2, 2027. Applicability and exceptions matter: these are not identical obligations for every AI application, and no GuardianChain compliance claim follows. DETAILS Duties and systemic-risk distinction (opens in new tab)Scope and application timetable (opens in new tab) Standards for agents and key systems These sources discuss safe links between agents and how to manage risk in key systems. NIST concept note NIST is developing an AI RMF profile on trustworthy AI in critical infrastructure. Its April 2026 concept note describes risk-management guidance under development, not a certification scheme or evidence that GuardianChain meets a standard. DETAILS nist.gov (opens in new tab) NIST AI Agent Standards Initiative Launched in February 2026, the initiative focuses on agents capable of autonomous action, secure operation, interoperability, industry-led standards and open protocols. It does not establish a GuardianChain integration, endorsement or adoption metric. DETAILS Initiative and scope (opens in new tab) NIST CAISI: securing AI agent systems NIST's January 2026 request for information says AI-agent security challenges hinder adoption, including risks where model outputs interact with software systems. This is research and public consultation, not an adopted certification standard. DETAILS January 12, 2026 RFI announcement (opens in new tab) How we read these. The findings do not mean models routinely escape. They explain why teams need clear limits, human decisions, lasting evidence and independent review. Earlier AI-control sources were reviewed against publishers' documents on 2026-10-08. The agent-standards, agent-security, critical-infrastructure and GPAI-provider references were checked on 2026-10-15. These sources explain market pressure; they do not demonstrate GuardianChain's effectiveness, availability or legal sufficiency. Sources for these cases TIER_0 supplied the links below. We did not open them or check them on our own in this pass. For AI labs and Any institution, we list background sources rather than an incident. 01 The agent that deleted the database Incident-source references supplied by TIER_0 Sources reviewed: 2026-10-21. Not fetched or independently verified in this pass. eWeek — 'Catastrophic Failure': AI Agent Wipes Production Database, Then Lies About ItAI Incident Database — Replit AI agent entity record 02 The trade that shouldn't have happened Incident-source references supplied by TIER_0 Sources reviewed: 2026-10-21. Not fetched or independently verified in this pass. U.S. SEC press release 2013-222 — Knight Capital, August 1, 2012 orderMoffatt v. Air Canada, 2024 BCCRT 149 03 The twelve seconds before impact Incident-source references supplied by TIER_0 Sources reviewed: 2026-10-21. Not fetched or independently verified in this pass. TechCrunch — October 24, 2023 robotaxi permit suspensionSF Standard — October 24, 2023 robotaxi permit suspension 04 One trail from curb to carousel Incident-source references supplied by TIER_0 Sources reviewed: 2026-10-21. Not fetched or independently verified in this pass. CrowdStrike — Root Cause Analysis — Channel File 291, August 2024 05 Provenance for every autonomous action Background, no incident Sources reviewed: 2026-10-21. Existing background references; not fetched in this pass. OpenAI o1 System Card (Apollo Research)Anthropic, Agentic Misalignment in Summer 2026 (July 13, 2026)European Commission: GPAI-provider obligations 06 What did the AI actually say? Incident-source references supplied by TIER_0 Sources reviewed: 2026-10-21. Not fetched or independently verified in this pass. Akin Gump — FCC Declaratory Ruling on AI-generated voice robocalls, February 8, 2024 07 The AI that touched the chart Incident-source references supplied by TIER_0 Sources reviewed: 2026-10-21. Not fetched or independently verified in this pass. HHS OCR — Change Healthcare Cybersecurity Incident FAQs 08 Your logs plus our receipt Background, no incident Sources reviewed: 2026-10-21. Existing background references; not fetched in this pass. NIST AI Risk Management FrameworkNIST AI 600-1 Photo credits Photographs are licensed from Pexels (free commercial use; no endorsement implied) and illustrate settings, not customers or deployments. Pexels license (opens in new tab). Robotic manipulator close-upPhoto by Pavel Danilyuk on Pexels. Source 8439084 (opens in new tab) Industrial robot armPhoto by Diego Martinez on Pexels. Source 29320998 (opens in new tab) Engineers testing a robotPhoto by ThisIsEngineering on Pexels. Source 3913031 (opens in new tab) Researcher controlling a manipulatorPhoto by Pavel Danilyuk on Pexels. Source 8438990 (opens in new tab) Wheeled delivery robotPhoto by Kindel Media on Pexels. Source 8566618 (opens in new tab) Bring us one real handoff. Pick one piece of AI-assisted work where the receiving side currently relies on summaries, screenshots or scattered logs. We will map it end to end and tell you plainly where GuardianChain helps and where it does not. Apply as a design partnerReview candidate use cases --- # Company | GuardianChain Source: https://guardianchain.org/company Last modified: 2026-10-24T11:51:00-05:00 Company Building accountability into AI. GuardianChain LLC builds the record that shows who approved an AI to act, what it was allowed to do, and what happened — so people stay accountable for the machines they use. HomeCompany Illustrative architecture, not a customer deployment. How we work How we work. Plain claimsWe say what is built, what is being built, and what this website does and does not do. Decisions stay humanWe record approval and evidence. Acceptance stays with the people responsible. Vendor-neutralWe work alongside your identity, workflow and security tools rather than replacing them. Verify, don't trustEvery record we produce can be checked by someone who does not trust us. We are at the design-partner stage and make no claims of production readiness, certification, customers or partnerships. Contact usResearch and sources Bring us one real handoff. Pick one piece of AI-assisted work where the receiving side currently relies on summaries, screenshots or scattered logs. We will map it end to end and tell you plainly where GuardianChain helps and where it does not. Apply as a design partnerReview candidate use cases --- # Design Partners | GuardianChain Source: https://guardianchain.org/design-partners Last modified: 2026-10-24T11:51:00-05:00 Design partners Shape the next stage with us. A small, honest collaboration. We learn from your real work; you shape what gets built. No platform access, commitments or guarantees are implied. Go to the applicationQuestions first HomeDesign Partners How it works A simple path, and you can stop at any step. Tell us what you're buildingHow your team uses AI and where you need clearer evidence. Explore the fitOne conversation about your priorities and one real workflow. Define the next stepA clear focus and shared expectations, or an honest 'not yet'. What a design partnership is not It is not platform access, a commercial commitment, a certification or a guarantee. Both sides can decide not to continue. Who we are looking for now: teams accepting AI-produced work; regulated operators; software vendors whose customers want provable governance. Who we are looking for Who this is for. Teams delivering agent work Teams delivering or receiving AI-assisted or multi-tool work Physical AI Vehicles, drones, robots Regulated operators Groups handing off regulated or sensitive research Who we are looking for Teams delivering or receiving AI-assisted or multi-tool work Operators who reconcile disputed or exceptional cases Groups handing off regulated or sensitive research What we would do together Map one real handoff end to end Define authority scopes and evidence expectations Test what a recipient needs to inspect What you receive Direct access to the people building it Influence over scope and priorities A written summary of what we learn about your workflow What we want to learn Which handoffs are most costly today What a recipient will and will not trust Where this does not fit Apply Tell us about your project. We reply honestly about fit. Please do not include secrets or sensitive evidence. Privacy notice. Common questions What does applying actually do? It sends your description to the GuardianChain team so we can evaluate fit and reply. It is stored only to respond and evaluate partnerships. It does not create an account, a contract or a commitment on either side. Will I get access to the GuardianChain platform? No. This site explains GC-1, GuardianBridge and FactReceipt but does not provide access to their runtimes. Applying and Partner Access never grant platform access. Explore the product architecture What should I include, and what should I leave out? A short, plain description of one delegation, handoff or reconciliation problem is enough. Do not send secrets, credentials, sensitive evidence or regulated personal information. Is there a guaranteed reply time or outcome? No. We reply honestly about fit, including when it is not a fit, but there is no guaranteed timeline or outcome. This site does not send email or an email receipt; the form confirms submission on screen. Can I see what I submitted later? Only if you were signed in when you submitted. Partner Access shows inquiries submitted while signed in; anonymous inquiries are not matched to your account by email address. How do I ask for my information to be deleted? Use the deletion or withdrawal request. It is queued for owner review; no records change until an action is confirmed. Application Name Email Organization What work would you like evidence and authority around? I have read the privacy notice (opens in a new tab; your form stays here) and agree that this information is stored to respond to me. I will not include secrets or sensitive evidence. Enable JavaScript to submit this form securely. Your details are not submitted without it. Built on established infrastructure What GuardianChain runs on. 48 services · 46 official marks available · 2 awaiting official artwork Railway Neon PostgreSQL Upstash Redis Stripe Cloudflare R2 AWS S3 Backblaze B2 Pinata (IPFS) Arweave Web3.Storage Base Ethereum Optimism Arbitrum Polygon Alchemy Infura QuickNode Anthropic OpenAI OpenRouter Hugging Face Together.ai Mistral xAI Stability AI Cohere Cloudflare AI Gateway Sentry PostHog Postmark Slack Cloudinary GitHub GitHub Actions npm Cloudflare 1Password Resend Google Workspace Notion Linear Figma Playwright Vite Cloudflare Hyperdrive TypeScript Node.js Names belong to their respective owners and are listed for identification only. No partnership, certification or endorsement is implied. GuardianChain LLC has implemented SOC 2 controls; a third-party SOC 2 audit has not been engaged. Bring us one real handoff. Pick one piece of AI-assisted work where the receiving side currently relies on summaries, screenshots or scattered logs. We will map it end to end and tell you plainly where GuardianChain helps and where it does not. Apply as a design partnerReview candidate use cases --- # Privacy | GuardianChain Source: https://guardianchain.org/privacy Last modified: 2026-10-24T11:51:00-05:00 Privacy Privacy notice Plain statements about what this temporary site does with your information. HomePrivacy Marketing status and identityInquiriesUpdatesPartner AccessStorage and protectionPublished-site analyticsCookies and browser storageRetention and requestsWhat happens next Last updated: 2026-10-23 Marketing status and identity Inquiry status history is stored in the marketing schema, separate from GuardianChain platform records. New submissions receive an initial received state; this is not evidence of human review or a service commitment. The sole owner is bootstrapped only from a matching, provider-verified sign-in email. The server then binds that account's OIDC issuer and subject; future owner checks use that exact pair. Email, client flags, profile fields and form values do not grant owner authority. Inquiries We store your name, email, organization or subject, message, submission type and time, and privacy acknowledgement in this application's marketing schema to respond to you and evaluate partner inquiries. This schema is separate from GuardianChain authority, evidence, receipt and operational records. One designated owner may review contact and design-partner inquiries, update their status, assign follow-up and add private internal notes. Those notes are not shown in Partner Access or sent to submitters. No evidence engine, authority service, billing service or marketing-email delivery is activated. Updates The separate product-updates form records your email, explicit consent, consent-notice version and submission time. Signing up records interest; it does not start email delivery or a paid subscription. Email delivery may be introduced later only with appropriate consent and unsubscribe controls. Partner Access Signing in uses an external identity provider to supply your identifier, email and available profile information. We store profile and session data and use a secure, HTTP-only session cookie for this limited site. Only inquiries submitted while signed in are linked to your account; anonymous inquiries are not claimed by matching email addresses. Signing in never grants operating-platform access. Storage and protection The site uses Replit-hosted application and database services. Forms are validated and protected with a honeypot and a submission limit. A time-bucketed hash of a network address is used for rate limiting; request diagnostics may also be processed by the hosting provider. Please do not submit secrets, sensitive evidence or regulated personal information. Published-site analytics The published informational site loads a Replit-hosted analytics script from i.replit.com/script.js. Checks of the eight public pages on October 15, 2026 (UTC) observed page-visit requests to i.replit.com/api/send containing the page URL, page title, hostname, referrer, screen dimensions, browser language and a site identifier. These observations cover signed-out public-page visits, not Partner Access sign-in or form submissions. Provider-side processing and retention were not established by this browser inspection. Cookies and browser storage Those checks observed a cookie named GAESA on guardianchain.org; Replit documents it as an infrastructure-routing cookie. No additional cookies, advertising requests, or localStorage or sessionStorage entries were observed during those public-page visits. This observation is not a guarantee about every browser configuration or future hosting change. Partner Access uses the separate sign-in and session cookies described above. No new analytics service has been added to this temporary site. Retention and requests We keep inquiries for 24 months after our last contact with you, then delete them, unless you ask us to delete them sooner. Use the dedicated request form to ask for deletion or to withdraw product-updates consent. A request stays in a pending queue until the configured owner reviews it; the email you enter is not proof of identity, and the owner must explicitly confirm any action. This site does not send email or an email receipt; the form confirms submission on screen. The pending request stores your email, request type and submission time, and its details are removed when the owner closes it. What happens next If a deletion request is confirmed, all contact, design-partner and updates submissions matching the request email are deleted from the active marketing data store, along with their marketing-only notes, review and status-history records. The pending request records for that email are removed too. This workflow does not inspect or erase infrastructure backup copies. It does not delete Partner Access profiles or sessions and never touches GuardianChain authority, evidence, receipts or operating records. If product-updates consent is withdrawn, the updates signup is retained with consent turned off; contact and design-partner inquiries remain unless deletion is separately requested. If a request is closed without action, no inquiry or consent data is changed. Completed request details are not retained. Request deletion or withdrawal Bring us one real handoff. Pick one piece of AI-assisted work where the receiving side currently relies on summaries, screenshots or scattered logs. We will map it end to end and tell you plainly where GuardianChain helps and where it does not. Apply as a design partnerReview candidate use cases --- # Contact | GuardianChain Source: https://guardianchain.org/contact Last modified: 2026-10-24T11:51:00-05:00 Contact Let's talk about AI governance. Questions about GuardianChain, a workflow you're evaluating, or working together? Tell us what you have in mind. HomeContact Ask a question. Name Email Subject Message I have read the privacy notice (opens in a new tab; your form stays here) and agree that this information is stored to respond to me. I will not include secrets or sensitive evidence. Enable JavaScript to submit this form securely. Your details are not submitted without it. Press and investor inquiries Contact Your data. Your choice. Deletion or consent withdrawal Submitting only queues a request. No records change until the owner reviews and explicitly confirms an action. RequestDelete my marketing inquiries and updates signupWithdraw consent for product updates Email used for your marketing submissions This creates a pending request only. The owner reviews it and must explicitly confirm any action. The email you enter is not proof of identity. No emails are sent. Enable JavaScript to submit this request securely. Your details are not submitted without it. Bring us one real handoff. Pick one piece of AI-assisted work where the receiving side currently relies on summaries, screenshots or scattered logs. We will map it end to end and tell you plainly where GuardianChain helps and where it does not. Apply as a design partnerReview candidate use cases --- # Terms and Conditions | GuardianChain Source: https://guardianchain.org/terms Last modified: 2026-10-24T11:51:00-05:00 Website terms Terms and Conditions Terms for accessing and using the GuardianChain website. HomeTerms and Conditions About these termsWebsite informationPermitted useContent and intellectual propertyInquiries and communicationsPrivacyLinks to other websitesAvailability and responsibilityChanges and questions Last updated: October 18, 2026 About these terms These terms describe permitted use of the GuardianChain website and its content. They apply to this informational website. Any access to GuardianChain products, partner programs or services is subject to a separate agreement. Website information This website describes GuardianChain’s approach, architecture and potential applications. Descriptions, examples and illustrations do not establish a service commitment or guarantee a particular outcome. Information may change as the technology and our work develop. Website content is general information and is not legal, regulatory or other professional advice. Evaluate its suitability for your circumstances before relying on it. Permitted use You may browse the website, share links to it and use its public information to evaluate GuardianChain. Do not use the website unlawfully, interfere with its operation, introduce malicious software, impersonate another person or attempt to access restricted systems without authorization. Content and intellectual property GuardianChain content, branding and original materials are protected by applicable intellectual property laws. Access to this website does not grant a license to GuardianChain technology or patents. Permission is required for uses of our materials beyond those allowed by law. Third-party photographs, logos, names and other materials remain subject to their owners’ rights and applicable licenses. Their appearance does not by itself establish a partnership or endorsement. Inquiries and communications An inquiry or expression of interest does not create a customer relationship, reserve access or commit either party to a project. Do not include passwords, confidential records or sensitive evidence in communications unless a suitable channel and agreement have been established. Privacy Read our Privacy Policy for information about personal information and privacy requests. Links to other websites External links are provided for reference. We do not control the content, availability or practices of third-party websites. Their own terms and privacy policies apply when you visit them. Availability and responsibility We aim to keep this website useful and accurate, but do not guarantee that it will always be available, current or free from errors. To the extent permitted by applicable law, the website and its information are provided as available without additional warranties. Nothing in these terms excludes rights or responsibilities that cannot lawfully be excluded. Changes and questions We may update these terms as the website changes. The date above identifies the latest revision. For questions about these terms, visit our Contact page. Bring us one real handoff. Pick one piece of AI-assisted work where the receiving side currently relies on summaries, screenshots or scattered logs. We will map it end to end and tell you plainly where GuardianChain helps and where it does not. Apply as a design partnerReview candidate use cases --- # Integrate GC-1 | GuardianChain Source: https://guardianchain.org/build Last modified: 2026-10-24T11:51:00-05:00 Build with GuardianChain Integrate GC-1. GC-1 connects to the systems you already use. Choose the way in that fits your team. HomeBuild Integration pathConnections THE 30-SECOND VERSION The problem AI agents now do real work inside companies: they read records, send messages, write code, move money. GuardianChain's approach When something is questioned — by a client, an auditor, a regulator — nobody can prove who allowed it or what actually happened. The only records live with the vendor, and they can be changed. What GuardianChain does Every AI agent gets a passport, approved by a named person. GuardianChain's approach Everything it does — and everything it is stopped from doing — becomes a FactReceipt: a sealed, timestamped record anyone can check, that no one can alter, and that stays yours even if the vendor disappears. Why it's different Your vendors' logs prove things to them. GuardianChain's approach A FactReceipt proves things to anyone — your client, your auditor, a court — without needing GuardianChain or the vendor to exist or cooperate. Why now Rules are arriving that require exactly this record: the EU AI Act's logging duties, state algorithmic-accountability laws, examiner expectations in finance. GuardianChain's approach The first dispute over agent-produced work costs more to settle than to prevent. What early partners gain Design partners work directly with the people building it, shape what gets built around their real workflows, and are first to hand their clients proof instead of promises. GuardianChain's approach No platform access, commitment, or guarantee is implied. Swipe to explore. Read the full picture ↓ Integration path From setup to your first check. 01Install Use a client library to connect an application to the same evidence record. 02Connect GC-1 connects to the systems you already use. Choose the way in that fits your team. 03First receipt Connect software that records events, resolves receipts and checks evidence. 04Verify Anyone can check a receipt. Choose your connection Fit the record to your workflow. These are integration paths to discuss with design partners. This website does not provide tools, keys or platform access. MCP Connection 01Connect an AI agent to the governance and verification tools. Acting with authority requires its human-bound passport.How MCP works This describes the connection; it is not an executable command. connector: MCP; passport: required for authority API / webhooks Connection 02Connect software that records events, resolves receipts and checks evidence.How API / webhooks works SDKs Use a client library to connect an application to the same evidence record. This describes the connection; it is not an executable command. connector: API; event: an action or outcomeconnector: SDK; record: linked evidenceconnector: webhook; source: your system Connectors Connection 03Let existing systems send events for signature and replay checks at the source.How Connectors works This describes the connection; it is not an executable command. connector: webhook; source: your system CLI and offline verification Connection 04Check an evidence bundle from the command line without depending on a live GuardianChain server.How CLI and offline verification works This describes the connection; it is not an executable command. input: evidence bundle; output: verification result Available to design partners. Design partners Verification is open. Anyone can check a receipt. Acting with authority requires a passport bound to an accountable person; an unbound agent may only witness itself. The connections you include define what is covered. Your evidence bundle states that coverage. Built on established infrastructure Everything we run on 48 services · 46 official marks available · 2 awaiting official artwork Runtime & data RailwayApplication runtime Neon PostgreSQLPostgreSQL database Upstash RedisRedis cache StripePayments Storage & permanence Cloudflare R2Object storage AWS S3Object storage Backblaze B2Object storage Pinata (IPFS)IPFS storage ArweaveStorage + anchoring Web3.StorageIPFS storage Anchor networks & RPC BaseNetwork EthereumNetwork OptimismNetwork ArbitrumNetwork PolygonNetwork AlchemyBlockchain RPC InfuraBlockchain RPC QuickNodeBlockchain RPC AI models & gateways AnthropicAI models OpenAIAI models OpenRouterModel gateway Hugging FaceModel hub Together.aiModel inference MistralAI models xAIAI models Stability AIGenerative media CohereAI models Cloudflare AI GatewayModel gateway Observability & communications SentryError monitoring PostHogProduct analytics PostmarkTransactional email SlackTeam messaging CloudinaryMedia management Build, security & operations GitHubSource hosting GitHub ActionsBuild automation npmPackage registry CloudflareEdge infrastructure 1PasswordSecrets management ResendTransactional email NotionWorkspace documents LinearIssue tracking FigmaInterface design ViteFrontend build Cloudflare HyperdriveDatabase acceleration TypeScriptType-safe development Node.jsJavaScript runtime Public RPC nodes Names belong to their respective owners and are listed for identification only. No partnership, certification or endorsement is implied. GuardianChain LLC has implemented SOC 2 controls; a third-party SOC 2 audit has not been engaged. Tell us what you want to connect. Bring one workflow and the systems it crosses. We will discuss where the record could help and where it cannot. Apply as a design partnerAsk a question --- # How to read a FactReceipt | GuardianChain Source: https://guardianchain.org/verify Last modified: 2026-10-24T11:51:00-05:00 Independent verification How to read a FactReceipt. A FactReceipt is a sealed record of one AI action: who approved it, what it was allowed to do, what it did, and where the proof lives. Here is how to read one and check it yourself. HomeVerify The receiptPublic proofVerify itYour content Request A person asks an agent to do something, within limits they set. Agent passport The agent presents its passport. No passport, no access. FactReceipt The sealed record: who allowed it, what was done, what was refused. Retrieve & verify Anyone can check the receipt. No account, no permission from us. FactReceipt Evidence anchored Retrieve & verify Anatomy of the record What a FactReceipt contains. A signed approval from a named person — we call it a grant. FactReceipt · Illustrative example fr_161761c9840bed32 action.refusedSealed · 4/6 anchored Illustrative ID only Whopp_7Q2K… · parent pp_3M9A… · lineage depth 1Grant gr_A41C… · Delivery owner (role) The passport, parent and human grant identify the agent and the role that allowed it to act. What was allowedread · write:staging in repo:acme/app · db:stagingDenied: write:production · limit 50 writes per hour2026-10-14/2026-10-28 · pol_v12 · sha256:5d1e… Scope, limits, expiry and the policy version describe the bounds of the approval. What happeneddatabase.execute → db:productionRequest sha256:a9f3… · source agent-runtimePrevious receipt sha256:77be… The request fingerprint, source and previous receipt connect this event to the surrounding record. When2026-10-21T14:02:11ZRFC 3161 token · tsa_ref …no later than block 52,922,083 (Base) Source time, timestamp evidence and a chain bound are different time references; the external values here are illustrative. What resultednone — refused before execution The stated effect is separate from the request; this illustrative action was refused before execution. What did not happenreport every 15 min · met An obligation and its recorded status show what was expected during the stated interval. What was preventedREFUSED rule 17 — no production writes during freezeMode: enforce The refusal names the attempted action and the rule in force; a refusal is evidence of restraint, not an executed write. Where the proof lives4 of 6 delivered · 2 pending (provider, awaiting batch window)Base · deliveredEthereum · deliveredOptimism · deliveredArbitrum · deliveredPolygon · pending — providerArweave · pending — awaiting batch windowStorageCloudflare R2AWS S3Backblaze B2IPFS (Pinata)ArweaveBatch root sha256:… · leaf 418 · path length 12 Delivered and pending anchors are stated separately; the transaction and Merkle values remain illustrative placeholders. Custody and retentionthe institution's own systemRetained until 2033-10-21 · last re-attested 2026-10-21 The custody and retention fields describe where content is kept and the date last recorded for its retention check. Who watches the watcherGC-1 build sha256:08c58dd6324ec9ae8116834449b5b42379da38d67cbbb16996eb16175239293cPolicy bundle sha256:8acd41da7ed8df10c6cbd9a87464ecbd449d91dcc5e42ed2ebc4870661e38159 The watcher fields identify the software and policy references recorded by this illustrative receipt. Canonicalisation: JCS (RFC 8785) · Hash: SHA-256 · Signature: Ed25519 { "schema": "guardianchain.factreceipt/v1", "receipt_id": "fr_161761c9840bed32", "type": "action.refused", "state": { "sealed": true, "anchors_owed": 6, "anchors_delivered": 4, "anchors_pending": 2, "pending_reason": "provider" }, "subject": { "agent_passport": "pp_7Q2K…", "lineage": { "parent_passport": "pp_3M9A…", "depth": 1 }, "human_grant": { "grant_id": "gr_A41C…", "signer": "Delivery owner (role)", "signer_key_fpr": "ed25519:9f2c…41ab", "sig_alg": "Ed25519", "issued": "2026-10-14T09:00:00Z", "expires": "2026-10-28T09:00:00Z" } }, "authority": { "scope": { "systems": [ "repo:acme/app", "db:staging" ], "actions": [ "read", "write:staging" ], "denied": [ "write:production" ] }, "limits": { "writes_per_hour": 50, "window": "2026-10-14/2026-10-28" }, "policy_version": "pol_v12 · sha256:5d1e…" }, "action": { "tool": "database.execute", "request_hash": "sha256:a9f3…", "target": "db:production", "source_system": "agent-runtime", "prev_receipt": "sha256:77be…" }, "restraint": { "decision": "REFUSED", "rule": "rule 17 — no production writes during freeze", "mode": "enforce" }, "time": { "source_time": "2026-10-21T14:02:11Z", "timestamp_authority": "RFC 3161 token · tsa_ref …", "chain_bound": "no later than block 52,922,083 (Base)" }, "result": { "effect": "none — refused before execution" }, "absence": { "obligation": "report every 15 min", "status": "met" }, "anchors": [ { "chain": "Base", "status": "delivered", "tx": "illustrative-transaction-01", "block": 52922083 }, { "chain": "Ethereum", "status": "delivered", "tx": "illustrative-transaction-02", "block": "…" }, { "chain": "Optimism", "status": "delivered" }, { "chain": "Arbitrum", "status": "delivered" }, { "chain": "Polygon", "status": "pending", "reason": "provider" }, { "chain": "Arweave", "status": "pending", "reason": "awaiting batch window" } ], "batch": { "merkle_root": "sha256:…", "leaf_index": 418, "inclusion_path_len": 12 }, "storage": [ "Cloudflare R2", "AWS S3", "Backblaze B2", "IPFS (Pinata)", "Arweave" ], "custody": { "content_location": "the institution's own system", "retention_until": "2033-10-21", "last_reattested": "2026-10-21" }, "watcher": { "gc1_build_digest": "sha256:08c58dd6324ec9ae8116834449b5b42379da38d67cbbb16996eb16175239293c", "policy_bundle": "sha256:8acd41da7ed8df10c6cbd9a87464ecbd449d91dcc5e42ed2ebc4870661e38159" }, "content_hash": "sha256:161761c9840bed32f275d0db2c7f0f7679388423d66d51621d0764ca38f346ad", "signature": { "alg": "Ed25519", "key_fpr": "ed25519:c29029a318a728e964bdd968910b573d27e0b8add9b36cc2a651c77c08f490ac", "value": "IS1a63iGxoxERI8jPY7wcHD-3kh6WszCfywXW3cdHrnT4-cZgET1aCj0Z8WPUaBb5EV5Tk4T6Kna3HA5DjIjCw" } } Canonicalise the receipt with JCS, omitting receipt_id, content_hash and signature. Compute SHA-256 and compare it with content_hash. Verify the Ed25519 signature over the 32 raw hash bytes with the published demo public key. Check the Merkle inclusion path against merkle_root. Look up the anchor transaction on the chain’s public explorer and compare the anchored root. Confirm the retention re-attestation date. Merkle, timestamp, transaction and retention values are illustrative; this page supplies a checkable demo hash and signature. Demo public key fingerprint: ed25519:c29029a318a728e964bdd968910b573d27e0b8add9b36cc2a651c77c08f490ac Download demo public key (PEM)Download demo key details Published demo public key (JWK) { "kty": "OKP", "crv": "Ed25519", "x": "zhdqRd_bs-glJxd9o-DKGkjXrXI0Z8opcFgoTe6XImw" } Illustrative command — not an installed tool or a live verification result. $ gc1 verify receipt.json hash: SHA-256 · signature: Ed25519 (demo key) hash input: omit receipt_id, content_hash, signature external proofs: illustrative placeholders Illustrative example. Values are generated for this page, signed with a demo key, and are not a GuardianChain production record. Illustrative record, not a receipt issued by this website. Public proof Where the proof lives. The evidence bundle carries the record and public proof needed to check it elsewhere. Anchored on public blockchains, stored with independent providers, and re-checked so you can prove the content was kept. One FactReceipt, six public blockchain anchors and five independent storage copies. Six public blockchains · five storage providers Three steps Verify it yourself in three steps. 01Open the receipt. DETAILS Read the approval, scope, event and the evidence behind it. 02Open its public proof. DETAILS Follow the bundle's anchor and verification instructions. 03Compare the fingerprints. DETAILS Check that the record and proof agree. Assess the underlying work separately. Verify without GuardianChain. The bundle holds the record and public proof so you can check it elsewhere. Verification can show that a record is intact and consistent. It does not decide whether the work is complete, suitable, or legally sufficient. Verification can show: The record is intact, consistent, and its evidence is where it says. Verification cannot show: The work is complete, fit for your purpose, or legally compliant. You may: Accept, hold and ask for more, or challenge. A FactReceipt covers what crossed the connectors GC-1 was given. Coverage is stated in every evidence bundle. Do I need a GuardianChain account? Public verification does not require an account or GuardianChain's servers. Does a passing check mean I should accept the work? No. The receiving party decides whether the work is complete and suitable for their purpose. Does a receipt recreate missing content? No. The content remains in the systems that hold it; retention records support checking that it was kept. Public verification tools are available to design partners. This website does not run verification or issue receipts. Non-custodial We never see your content. Only a fingerprint leaves your systems. Your content stays with you. A document remains inside your building. Only a fingerprint leaves across the system boundary. Only a fingerprint crosses the boundary; your content never leaves your systems. Bring a record you need to understand. Tell us what your team needs to check and which evidence is difficult to carry across systems. Apply as a design partnerSee connection options