A doctor takes the pulse before reading the chart. Heartbeat Attested™™ is the pulse layer for one AI deployment. It signs every output. Each governed decision produces a signed receipt. The receipts hash-chain into an audit ledger that any auditor can verify without contacting us. When the AI behaves correctly, the pulse is steady. When it drifts, the pulse changes, the receipts show it, the ledger keeps the record. Runs where the data lives. Never pings out.
Audio plays muted by default. Click the speaker icon to hear the 60 BPM pulse.
A worked example for a mid-sized AI vendor with European customers. Replace the revenue figure with your own. The conclusion does not change.
Three percent of €150M in global annual turnover. Per violation. The EU AI Office can issue compounding penalties for ongoing non-compliance. Your audit logs sit on a server you administer, which means you cannot prove the logs are contemporaneous.
One-time prepay. Twelve months of continuously signed, hash-chained evidence streams that any auditor verifies with an open-source script. Producing defensible evidence costs a rounding error against the exposure you already carry. The license buys the evidence, not a guarantee about the fine.
Three real cases from public court records and regulatory action. Each one shows what happens when AI is deployed without verifiable governance evidence, and what Heartbeat Attested™ would have done if it had been in the chain.
A federal court held that attorney-client privilege can be waived when confidential client data passes through third-party AI servers. The defending firm could not produce evidence that the AI use was governed, scoped, or contained. The privilege determination turned on what could not be proven about the AI interaction.
An AI-driven claims-decision system has been alleged in litigation to have produced denials at scale with a high error rate, with downstream harm to patients. The complaint cites the absence of a verifiable audit artifact for each AI-assisted decision and the difficulty of reconstructing what the system did at the moment of each denial.
Italy's data protection authority issued an emergency processing ban on Replika in February 2023 after determining it failed to protect minors from harmful content. No verifiable evidence of age-adaptive governance existed at the time of the regulator's review.
The architecture does not care what the AI is doing. It cares that every governed output produces an evidence beat. Five buyers, five regulatory anchors, one license shape.
Three steps, one architecture. The kernel runs where the data lives. It never pings out. The receipt is produced as a by-product of governance, not a report someone writes after the fact.
docker compose up deploys the governance kernel inside your environment, bound to your license file. Every AI output passes through it. The decision (allow, refuse, halt, escalate) is made on local computation alone. The capability is local or it does not run in that session.
Each governance event produces a cryptographically signed receipt, hash-chained to the previous one. Tamper-evident, append-only, signed on your machine with material generated there. Nothing pings out. Nothing leaves. The receipt is real because nothing ever did.
Export the bundle. Your auditor, counsel, or regulator runs the open-source verifier against your public key. The chain either matches or it doesn't. Verification is a property of the cryptography. You never call us. We never see your data. Proof you can hold, made where the data lives, sent nowhere.
The verifier is open-source under the MIT License. Python standard library plus the cryptography package. No network calls. No telemetry. The same script that ships inside every evidence package is also public, so an auditor can fetch it independently and confirm byte-identical.
The same Heartbeat Attested™ evidence stream is responsive to the regulatory cascade behind Article 50. The lite tier is Article 50-scoped. Multi-framework coverage lives in the Enterprise tier.
The unit is one AI surface, one stable identifier, like one major legal-AI deployment, one major productivity-suite AI tenant, one foundation-model enterprise endpoint, one internal RAG. The rung is set by the number of attested operators inside that surface. Locate yourself; the rung that fits is the rung you discuss with us.
The license buys the evidence, not a guarantee about the fine.
Procurement and security review run on their own timeline regardless of where a buyer starts on the ladder. The conversation is where the verifier is demonstrated live against your environment, the license language is reviewed, and the right rung is confirmed.
Twelve regulation-specific deliverable packs. Each one attaches to any Heartbeat Attested™ tier. The premium adds to your base tier price. Every bundle ships with the standing disclosure.
This guide helps you prepare for the named regulation. It does not certify your compliance. You remain responsible for your specific deployment. Where regulatory ambiguity exists or your situation has facts that affect the legal analysis, consult your own counsel for your specific deployment. Heartbeat Attested™ is infrastructure, not a certifier.
Covers AI systems interacting with natural persons, generating synthetic content, or performing emotion or biometric categorisation.
For Annex III high-risk AI systems across the regulated categories.
High-risk AI making decisions in employment, education, financial services, healthcare, housing, insurance, and essential government services.
NYC employers and employment agencies using automated employment decision tools must conduct annual independent bias audits and notify candidates.
California requires generative AI developers serving California persons to post training-data documentation.
Insurers using AI in underwriting, rating, marketing, fraud detection, and claims must maintain a written AI Systems Program.
US banks deploying AI or ML models for credit, market, operational, compliance, or capital decisions must manage model risk across the lifecycle.
FDA-regulated entities using AI in workflows producing records subject to FDA inspection.
EU pharma manufacturers using AI in GMP-regulated computerised systems.
Manufacturers of AI or ML-enabled medical devices applying the GMLP Guiding Principles.
Federal contractors and federal AI buyers requiring alignment to the NIST AI RMF and the Generative AI Profile.
Watchlist bundle. Texas Responsible AI Governance Act is pending in the legislature. Preparation framework mapped to the most recent publicly available bill text; any enacted statute supersedes.
This product does one thing completely: it proves one AI deployment is governed. Read both columns so the boundary is clear before you buy.
One AI surface is one stable identifier: one major legal-AI deployment, one major productivity-suite AI tenant, one foundation-model enterprise endpoint, one internal RAG, one customer-facing chatbot, one drug-discovery AI assistant, one semiconductor-traceability AI. The container kernel binds to that single deployment identifier and signs every governance event from it. If your organization runs three separate AI surfaces, you need three licenses. If one surface serves a million queries, you still need one license. The unit is AI surfaces, not seats or traffic. Tier inside the license is set by the number of distinct attested operators using that surface.
No. Heartbeat Attested™ produces cryptographic evidence for one bound AI deployment. That evidence is what an auditor or regulator needs to verify governance of that deployment. It is not, on its own, compliance for your organization. Compliance is a posture that covers every AI surface in your estate, every framework that applies to you, and every operational control you run. Heartbeat is one piece of evidence in that picture. If you have multiple AI surfaces, drift exposure, fleet operations, or multi-framework requirements, those need Regulayer™.
No, and that distinction protects you. We are infrastructure. You generate signed evidence, you and your auditor interpret it, you and your regulator settle it. We do not certify your compliance, which means there is no certifier in your chain whose opinion can be challenged in court or whose insurance can be invalidated. You hold the keys. You hold the evidence. We hold the patent and the verifier, neither of which we can use against you.
The pulse changes. The receipts record the drift. The ledger preserves the record. You hold the evidence. The product does not decide what drift means; that is for your team, your auditor, your counsel, or your regulator to interpret. The point of the architecture is that the evidence is there, contemporaneous, signed, and verifiable, before anyone asks for it.
The software, the documentation pack, the install guide, the audit-defense playbook, and the open-source verifier, all delivered by email at purchase. The container is built to self-install via a single Docker Compose command. Error messages include remediation steps. The verifier ships with sample receipts and a verification walkthrough. Beyond that, the license does not bundle a support contract, an SLA, or consulting hours; those are commercially separate engagements available when an organization wants them.
Those companies sell policy dashboards and checklist tooling. They help you prepare for and maintain certifications. None of them produces cryptographic evidence streams that a regulator can verify independently. Heartbeat is the substrate underneath that question. Different category, different price bracket, different defensibility. We compete with the absence of evidence infrastructure, not with the presence of dashboards.
Install the container. One command: docker compose -f docker-compose.container_kernel.yml up. The container is bound to your license file (which you receive by email after purchase) and exposes a single POST /govern endpoint your AI pipeline calls. Integration is one engineer-day for a team that already runs Docker. The public verifier confirms your integration is correct before you ever show evidence to a regulator.
No. The SDK runs entirely on your infrastructure. The signing key is generated on your machine on first run and never leaves it. There is no telemetry, no analytics, no remote update channel, no metrics endpoint. You can block all outbound network traffic from the SDK at your firewall and it continues to function. We do not know how many heartbeats you emit and we do not want to.
The SDK keeps running. New heartbeats stop being covered by the Heartbeat Attested mark license. Every receipt signed before expiry remains cryptographically valid and verifiable forever. A regulator looking at your historical record from the licensed period sees an unbroken stream. Letting the license lapse creates a gap going forward, not backward. That asymmetry is the renewal argument.
The May 2026 Omnibus extended high-risk system enforcement to December 2027 and embedded safety components to August 2028. Article 50 was preserved on the original timeline. If Article 50 itself moves, the same evidence stream is responsive to NIST AI RMF, FDA-EMA Principle 9, EU GMP Annex 11, ISO 42001, and the state-level US AI acts. Multi-framework coverage sits in the Enterprise tier. The infrastructure you bought for one cliff defends you under the next eleven.
Patent-pending technology developed by Regulayer, Inc. Counsel held under filing.
Heartbeat Attested covers one bound AI deployment. Regulayer covers an organization's AI estate. Heartbeat gives you a signed receipt stream, an independent verifier, and the container kernel. Regulayer adds fleet aggregation across all your AI surfaces, predictive drift monitoring, meta-governance, multi-jurisdictional certificate engines, model-evolution control, safe-mode fallback, and the broader claim set the patent portfolio covers. A buyer with multiple AI surfaces, drift exposure, or fleet requirements needs Regulayer; one bound deployment fits Heartbeat.