LiabilityLoop

Real-time attribution engine that maps autonomous agent decisions to human legal entities for insurance and compliance.

Insurers and regulators increasingly refuse coverage without granular, immutable proof of human oversight. LiabilityLoop issues cryptographic chain-of-custody certificates for every agent action — so liability always lands on a named legal entity, not a black-box model.

Primary buyer	Insurers · risk desks · regulated AI operators
Product	Attribution API + certificate issuance
Model	Per-transaction fee + annual platform license
Price	$0.05 / verified action + $50k platform fee
Moat	Path to regulatory certification as an AI liability standard

Why it matters

When agents execute trades, diagnoses, or policy changes, “who approved this?” must be answerable in milliseconds — with cryptographic evidence that survives audit, litigation, and reinsurer review.

Architecture

Agent telemetry and human-in-the-loop evidence converge in the attribution engine. Every decision exits as a signed chain-of-custody certificate that insurers, regulators, and internal audit can verify independently.

flowchart TB
  subgraph Sources["Decision sources"]
    Agent["Autonomous agent"]
    Approver["Human legal entity<br/>approver / overseer"]
    Framework["Agent framework<br/>plugins"]
  end

  subgraph Core["LiabilityLoop core"]
    Ingest["Telemetry ingest"]
    HITL["HITL evidence<br/>collector"]
    Engine["Attribution engine"]
    Certify["Certificate issuer<br/>digest + HMAC / KMS"]
    Store[("Immutable certificate store")]
  end

  subgraph Consumers["Consumers"]
    API["HTTPS API<br/>/v1/certificates"]
    Insurer["Insurer / reinsurer"]
    Regulator["Regulator / auditor"]
    Viewer["Certificate viewer"]
  end

  Agent -->|action + payload hash| Ingest
  Framework --> Ingest
  Approver -->|role · LEI · approval time| HITL
  Ingest --> Engine
  HITL --> Engine
  Engine -->|map agent → legal entity| Certify
  Certify --> Store
  Certify --> API
  API --> Insurer
  API --> Regulator
  API --> Viewer
  Store -.->|verify signature| Insurer

Loading

Design principles

• Human-in-the-loop required — actions without oversight evidence issue as incomplete, not silently valid
• Cryptographic custody — each certificate carries a digest and signature suitable for KMS/HSM upgrade
• Insurer-ready artifact — one verifiable object per agent action for claims and compliance workflows
• Framework-agnostic — ingest from any agent runtime via API or plugin

Quick start

npm install
npm test
npm run demo
npm run serve   # http://localhost:8787

Issue a certificate

curl -s http://localhost:8787/v1/certificates \
  -H 'content-type: application/json' \
  -d '{
    "actionId": "trade-88421",
    "agentId": "agent-exec-7",
    "description": "Submit equity order",
    "payloadHash": "a1b2c3",
    "occurredAt": 1893456000000,
    "oversight": [{
      "legalEntityId": "lei:CH-SWISSRE-RISK-01",
      "displayName": "Swiss Re Emerging Risks Desk",
      "role": "approver",
      "approvedAt": 1893455998800
    }]
  }'

SDK usage

import { AttributionEngine, verifyCertificate } from "@aifrontiers/liabilityloop";

const engine = new AttributionEngine({
  signingKey: process.env.LIABILITYLOOP_SIGNING_KEY,
  requireHumanInLoop: true,
});

const cert = engine.attribute({
  actionId: "trade-88421",
  agentId: "agent-exec-7",
  description: "Submit equity order",
  payloadHash: "a1b2c3",
  occurredAt: Date.now(),
  oversight: [/* human legal entities */],
});

console.log(cert.status, verifyCertificate(cert, process.env.LIABILITYLOOP_SIGNING_KEY!));

Repository layout

src/attribution/   Human-in-the-loop mapping engine
src/certificates/  Digest + signature chain-of-custody issuance
src/api/           HTTP API (/v1/certificates)
src/cli/           Certify / serve CLI
apps/demo/         Certificate viewer
docs/              Architecture notes

Roadmap focus

Plugin for a major agent framework that emits audit trails for human review — this repo ships the certificate core and HTTP API.

License

MIT