Enterprise Workflows

Real-world examples of QuantZK VDI in regulated industries.

Fair Lending (Financial Services)

Use Case

A bank's AI lending system must show cryptographic evidence that encoded fair-lending checks passed for each decision — without revealing the applicant's private data or the model's proprietary logic.

Minimal example (loan approval pipeline)

Generic VDI attest() for a lending decision (same shape as the protocol SDK, not the billing pilot):

import { attest, verify } from '@quantzk/attest';

const attestation = await attest({
  pipeline: {
    name: 'Loan Approval',
    manifest: 'fair-lending-v1',
    steps: ['application', 'identity', 'credit', 'fairness', 'pricing', 'decision']
  },
  decision: {
    path: ['application', 'identity', 'credit', 'fairness', 'pricing', 'decision'],
    inputs: { credit_score: 720, income: 85000, debt_ratio: 0.32 },
    outcome: 'decision'
  },
  authority: {
    type: 'legal_review',
    name: 'Example Counsel LLP',
    ref: 'Opinion #2026-AI-FL-014'
  }
});

const result = await verify(attestation);
console.log(result.valid); // true when the encoded relation and manifest checks pass

Private inputs (credit_score, income, debt_ratio) are not included in the attestation. The proof shows they satisfied the manifest constraints — not that the manifest perfectly captures every legal obligation (that assurance comes from the bound authority and external review).

Full pipeline

import { attest, verify } from '@quantzk/attest';

const attestation = await attest({
  pipeline: {
    name: 'ACME Bank Lending Pipeline',
    manifest: 'fair-lending-v1',
    version: '2.0.0',
    steps: [
      'application',       // Applicant submits data
      'identity',          // KYC/AML verification
      'credit_analysis',   // Credit scoring
      'fairness_check',    // Fair lending compliance
      'pricing',           // Rate determination
      'decision'           // Final approval/denial
    ]
  },
  decision: {
    path: ['application', 'identity', 'credit_analysis', 'fairness_check', 'pricing', 'decision'],
    inputs: {
      credit_score: 720,
      income: 85000,
      debt_ratio: 0.32,
      approved: true,
      roe_compliant: true
    },
    outcome: 'decision'
  },
  authority: {
    type: 'legal_review',
    name: 'Example Counsel LLP',
    ref: 'Legal Opinion #2026-AI-FL-014',
    regulation: 'Fair Lending Act, ECOA'
  }
});

Constraints Enforced

Constraint	Description	Critical
no_protected_class	Decision does not use race, gender, religion, national origin, marital status, age, or receipt of public assistance	Yes
adverse_action_justified	Any adverse action justified by legitimate credit risk factors per ECOA Section 701(a)	Yes
rate_within_bounds	Interest rate within regulatory bounds for applicant risk tier	No
equal_information_requirement	Same information requested from all applicants regardless of protected class	Yes

Audit Flow

1. Bank's lending AI generates an attestation for every loan decision
2. Internal compliance team runs verify() and emits signed receipts
3. External auditor (e.g., OCC examiner) receives attestations and receipts
4. Auditor verifies independently, no bank system access needed
5. Regulator receives cryptographic evidence that encoded checks passed — without seeing private applicant data (manifest authoring and legal mapping remain separately attestable)

---

Clinical AI (Healthcare)

Use Case

A hospital's AI diagnostic system must prove it handled PHI according to HIPAA rules without revealing the patient's medical records.

Pipeline

const attestation = await attest({
  pipeline: {
    name: 'Clinical Diagnostic Pipeline',
    manifest: 'hipaa-phi-v1',
    version: '1.0.0',
    steps: [
      'data_access',      // PHI retrieval
      'preprocessing',    // Data normalization
      'inference',        // AI model execution
      'logging',          // Audit trail
      'output'            // Result delivery
    ]
  },
  decision: {
    path: ['data_access', 'preprocessing', 'inference', 'logging', 'output'],
    inputs: {
      patient_id_hash: '0xabc...',
      data_fields_accessed: 3,
      purpose: 'treatment',
      output_de_identified: true,
      access_logged: true
    },
    outcome: 'output'
  },
  authority: {
    type: 'audit_firm',
    name: 'Example Healthcare Auditor',
    ref: 'HIPAA Audit Report #EXAMPLE-HC-042'
  }
});

Constraints Enforced

Constraint	Description	Critical
minimum_necessary	Only minimum necessary PHI accessed	Yes
authorized_purpose	Access for treatment, payment, or healthcare operations	Yes
audit_trail	Access logged with user, timestamp, and purpose	Yes
de_identification	Output does not contain identifiable PHI	Yes

Compliance Flow

1. Clinical AI generates attestation for each diagnostic run
2. Hospital's HIPAA compliance officer verifies attestations
3. During HHS audit, attestations provide cryptographic evidence that encoded HIPAA checks passed
4. Patient data never leaves the hospital, only proofs and receipts

---

EU AI Act (High-Risk AI Systems)

Use Case

An autonomous hiring system operating in the EU must prove it provides adequate human oversight as required by EU AI Act Article 14.

Pipeline

const attestation = await attest({
  pipeline: {
    name: 'EU Compliant Hiring System',
    manifest: 'eu-ai-act-art14-v1',
    version: '1.0.0',
    steps: [
      'intake',            // Application received
      'screening',         // Initial AI screening
      'oversight_check',   // Human oversight verification
      'monitoring',        // Anomaly detection
      'decision'           // Hire/reject
    ]
  },
  decision: {
    path: ['intake', 'screening', 'oversight_check', 'monitoring', 'decision'],
    inputs: {
      human_reviewer_present: true,
      reviewer_certified: true,
      anomaly_check_passed: true,
      intervention_available: true,
      bias_mitigation_active: true
    },
    outcome: 'decision'
  },
  authority: {
    type: 'standards_body',
    name: 'CEN-CENELEC JTC 21',
    ref: 'EN XXXXX:2026, AI Conformity Assessment'
  }
});

Constraints Enforced

Constraint	Description	Critical
human_oversight_mechanism	System has human oversight appropriate to risk	Yes
system_understanding	Human overseers understand system capacities and limitations	Yes
anomaly_detection	System supports detection of anomalies and dysfunctions	Yes
intervention_capability	Human can intervene or interrupt system operation	Yes
automation_bias_awareness	Measures to guard against automation bias	No

---

Autonomous Agent Networks

Use Case

A fleet of autonomous trading agents needs to verify each other's decisions without a central authority and without revealing proprietary strategies.

Three-Agent Trust Propagation

import { attest, verify, verifyAndReceipt, verifyReceipt } from '@quantzk/attest';
import { randomBytes } from 'crypto';

// Agent A: Trading Bot, makes a trade decision
const attestation = await attest({
  pipeline: {
    name: 'Algo Trading Pipeline',
    steps: ['market_data', 'signal_generation', 'risk_check', 'position_sizing', 'execution'],
    constraints: [
      { id: 'max_position_size', vertex: 'position_sizing', critical: true },
      { id: 'risk_limit', vertex: 'risk_check', critical: true },
      { id: 'market_hours', vertex: 'execution', critical: true }
    ]
  },
  decision: {
    path: ['market_data', 'signal_generation', 'risk_check', 'position_sizing', 'execution'],
    inputs: { signal_strength: 0.87, risk_score: 0.23, position_pct: 0.02 },
    outcome: 'execution'
  },
  authority: { type: 'self_attested' }
});

// Agent B: Risk Monitor, verifies and emits receipt
const { verification, receipt } = await verifyAndReceipt(attestation, {
  signingKey: randomBytes(32).toString('hex'),
  verifierId: 'vdi:verifier:risk-monitor-001'
});

// Agent C: Portfolio Manager, trusts Agent B's receipt
const receiptOk = await verifyReceipt(receipt, {
  attestation,
  trustedVerifiers: ['vdi:verifier:risk-monitor-001']
});
// receiptOk.valid === true, no ZK re-verification needed

Scaling Trust in Agent Networks

Operation	Time	Network Scale (1000 agents)
Full ZK verification	~27ms	27 seconds
Receipt verification	~2ms	2 seconds
Receipt propagation	~0.1ms	100ms

With verification receipts, trust scales linearly rather than exponentially. One full verification produces a receipt that 999 agents can trust without re-verifying.

---

NIST AI Risk Management Framework

Use Case

A government agency deploying AI must demonstrate compliance with NIST AI RMF (AI 100-1) for risk governance, context mapping, performance measurement, and risk management.

Pipeline

const attestation = await attest({
  pipeline: {
    name: 'Federal AI Deployment Pipeline',
    manifest: 'nist-ai-rmf-v1',
    version: '1.0.0',
    steps: [
      'governance',        // Risk management policies
      'documentation',     // System context and capabilities
      'evaluation',        // Performance measurement
      'risk_management',   // Risk prioritization
      'deployment'         // Go/no-go
    ]
  },
  decision: {
    path: ['governance', 'documentation', 'evaluation', 'risk_management', 'deployment'],
    inputs: {
      policies_in_place: true,
      capabilities_documented: true,
      metrics_measured: true,
      risks_managed: true
    },
    outcome: 'deployment'
  },
  authority: {
    type: 'standards_body',
    name: 'NIST',
    ref: 'AI 100-1 (January 2023)'
  }
});

---

Custom Constraints

You can define your own constraints for any industry or use case:

const attestation = await attest({
  pipeline: {
    name: 'Custom Content Moderation',
    steps: ['intake', 'classification', 'policy_check', 'action'],
    constraints: [
      { id: 'no_false_positives_on_news', vertex: 'classification', critical: true,
        description: 'News content must not be classified as misinformation without high confidence' },
      { id: 'human_review_threshold', vertex: 'policy_check', critical: true,
        description: 'Content with borderline scores must be routed to human review' },
      { id: 'appeal_mechanism', vertex: 'action', critical: false,
        description: 'User must be notified and provided an appeal mechanism' }
    ]
  },
  decision: {
    path: ['intake', 'classification', 'policy_check', 'action'],
    inputs: { content_hash: '0x...', confidence: 0.95, routed_to_human: false },
    outcome: 'action'
  },
  authority: { type: 'self_attested' }
});