# Hudson Data — machine-readable site overview

Last reviewed: 2026-07-21

## Brand

Hudson Data is the company building HD Trust — session-trust intelligence for the AI-fraud era — and the broader Centurion fabric around it. Decade of building AI/ML for credit, fraud, and trust at production scale inside leading US banks, carriers, and fintechs.

URL: https://hudsondata.com
Contact: info@hudsondata.com
SOC 2 Type 2 certified.

## The headline product: HD Trust

**Tagline:** Stop AI-driven fraud at the session — for the bank, and for the customer being targeted.

HD Trust is session-trust intelligence built for the AI-fraud era. Not a fingerprinting platform. Not a black-box ML score. A three-axis decisioning matrix that the customer controls.

### The three-axis model

Every session resolves into three classifications:

- **WHO** is coming to your system? — identity, behavior, history. Values: LEGIT_OWNER, COACHED, STOLEN_ID, MULE, BOT_SCRIPT, AI_AGENT.
- **WHAT** device are they using? — device authenticity. Values: AUTHENTIC, PRIVACY, COMPROMISED, HEADLESS, STEALTH, AI_BROWSER.
- **HOW** are they connecting? — connection integrity. Values: DIRECT, PRIVACY_VPN, PROXY, TAMPERED.

The three classifications resolve into a cell of the customer's decisioning matrix. The cell IS the action. Verdict actions: APPROVE, PROTECT, STEP_UP, REVIEW, BLOCK.

PROTECT exists because legitimate customers on compromised environments need help, not a block screen. Fingerprinting platforms cannot make this distinction because they only see the device.

### Underlying signal channels

The three classifications are fed by multiple session-signal channels:

- WHO signal: identity continuity, biometric familiarity, account-history coherence, behavioral cadence
- WHAT signal: hardware integrity, stealth-browser signatures, malware indicators, automation traces
- Network signal: TCP fingerprint, MTU anomalies, residential-proxy detection (physics-based), carrier-IP coherence
- Telephony signal: SIM-swap risk, line-type classification, carrier reputation
- Behavior signal: keystroke dynamics, mouse-movement patterns, form-fill cadence, hesitation patterns
- Resilience signal: SDK integrity, transport-layer tamper detection, MITM-proxy detection, payload-replay analysis

### Seven AI-era threats HD Trust catches

Threats where dimensional architecture is the difference vs. fingerprinting platforms:

1. **Voice-coercion fraud** (WHO): Scammer coaching the customer over the phone
2. **Remote-access scam / RAT** (WHO + WHAT): TeamViewer / AnyDesk on the customer's device
3. **AI-agent browser attack** (WHO): Claude / GPT controlling a real browser to apply for credit
4. **Stealth browser** (WHAT + HOW): Multilogin / GoLogin / AdsPower with coherent fingerprint
5. **Push-payment scam** (WHO): Real customer, real device, coerced into a fraudulent transfer. HD Trust catches behavioral signal; combined with bank's payment-context completes the detection.
6. **Malware-mediated theft** (WHAT + HOW): Banking trojan injecting form overlays
7. **MITM / proxy injection** (HOW): Adversary intercepting or replaying responses

### Decision output (API)

Every HD Trust decision returns a structured object:

- who_classification: enum (LEGIT_OWNER | COACHED | STOLEN_ID | MULE | BOT_SCRIPT | AI_AGENT)
- what_classification: enum (AUTHENTIC | PRIVACY | COMPROMISED | HEADLESS | STEALTH | AI_BROWSER)
- action: enum (APPROVE | PROTECT | STEP_UP | REVIEW | BLOCK) — read from the WHO × WHAT × HOW cell of the customer's matrix
- matrix_policy_version: string (versioned, replayable)
- dimension_evidence: per-dimension underlying signal
- reason_codes: top-K explainable signal drivers
- linked_sessions: consortium-aware session linkages
- trajectory: STABLE | IMPROVING | DEGRADING | COMPROMISED

### Integrations

- Web SDK (JS, 3-line install)
- Mobile SDKs (iOS, Android, React Native, Flutter, Capacitor)
- REST API
- gRPC / streaming
- Webhooks
- Dashboard (case management)
- Data-lake export (S3, GCS, Snowflake, BigQuery, Databricks)
- SIEM integration (Splunk, Sentinel, Chronicle, Elastic)

### Deployment

- Hudson-managed SaaS (multi-tenant)
- Customer-cloud, hybrid, and on-prem options are roadmap items unless explicitly agreed in an enterprise engagement
- Data residency and processing boundaries are documented for the contracted service configuration

### Regulatory alignment

- CFPB consumer-protection
- NIST AI Risk Management Framework
- OCC AI risk management
- NY DFS Part 500
- GDPR / CCPA support depends on customer use, notices, lawful basis, configuration, and jurisdiction
- SOC 2 Type 2

## AI-Fraud Audit

Three-week written assessment of the customer's session-trust posture against AI-era threats.

Phases:
1. Scoping conversation (30 min)
2. Posture assessment (your environment, your tooling, current coverage)
3. AI-era gap map (vs. threats fingerprinting can't see)
4. Written report + 60-min walk-through

Deliverables:
- Documented baseline of current posture
- Gap map vs. AI-era threats, with severity
- Prioritized 30/60/90-day roadmap (vendor-neutral)
- Regulatory alignment notes (CFPB, OCC, NIST AI RMF, NY DFS)

## Centurion Platform (the broader fabric)

HD Trust is the front. Centurion is what plugs in when the customer needs credit-and-fraud decisioning end-to-end.

- **Model Foundry**: Credit risk model development factory. Bureau-native (TransUnion, Experian, Equifax), audit-ready, forward-deployed. Builds models across the credit lifecycle: lead-scoring, origination, pricing, line-assignment, behavioral, collections, recovery, CECL/IFRS-9.
- **ML Graph**: Real-time graph engine for fraud and abuse, with graph features designed for use at decision time rather than only offline case review.
- **FlowX**: The unified decisioning fabric — versioned, A/B-testable, audit-logged. Where HD Trust scores, Model Foundry models, and ML Graph signals compose into one decision.

## Forward-deployed delivery

HD Trust and Centurion ship with a forward-deployed team that integrates the service into the customer's production stack, operates alongside the customer team, and stays accountable to agreed outcomes. HD Trust is Hudson-managed SaaS today; broader Centurion work may run in customer-controlled infrastructure when explicitly scoped.

Phases:
1. Audit (weeks 1-3)
2. Stand-up (weeks 4-10)
3. Tune & integrate (months 3-6)
4. Operate (ongoing)

Principles:
- Senior practitioners only — no junior pyramid
- A documented deployment model, data flow, and control boundary
- Operated, not delivered
- Accountable to outcomes (detection lift, false-positive reduction, customer-protection metrics)
- Audit and regulator-aware (SR 11-7, OCC AI, NIST AI RMF, CFPB)
- Hand-off as a deliverable

## Use cases

Three-lens organization:

**By customer journey** — five lifecycle stages:
1. Attract (prospecting & lead)
2. Acquire (apply & onboard)
3. Engage (authenticate & transact)
4. Grow (service & expand)
5. Resolve (collect & recover)

**By risk pillar:**
- Credit decisions: application, pricing, line, loan amount, CLI/CLD, behavioral, collections, recovery
- Fraud decisions: account opening, login, payment authorization, claims, promo, step-up, mule detection, manual-review prioritization

**By industry:**
- Fintech lender
- Card issuer
- Insurance carrier
- Bank
- Marketplace / E-commerce

Industry-specific deep dives:
- [/use-cases/bank](https://hudsondata.com/use-cases/bank)
- [/use-cases/insurance-carrier](https://hudsondata.com/use-cases/insurance-carrier)
- [/use-cases/marketplace-ecommerce](https://hudsondata.com/use-cases/marketplace-ecommerce)

## Proof points

Case studies (banking PIN POS fraud reduction, auto-claims fraud savings, fintech credit underwriting at portfolio scale) are shared with prospects under NDA, not posted publicly. Contact us at https://hudsondata.com/contact-us to request them.

## What HD Trust isn't

- Not a device fingerprinting tool
- Not a black-box ML score
- Not a CAPTCHA
- Not self-install SaaS

## Why HD Trust now

- AI-era fraud (deepfakes, AI agents, voice coercion, stealth browsers, remote-access scams) has moved from research papers to operator toolkits in the last 18 months
- Regulatory pressure converging: CFPB on push-payment-scam reimbursement, NIST AI RMF, OCC AI guidance, NY DFS consumer-fraud expectations
- Fingerprinting platforms architecturally cannot see what HD Trust sees: dimensional architecture (WHO × WHAT × HOW) catches threats single-axis systems miss

## Contact & next steps

- Talk to our team: https://hudsondata.com/contact-us
- Schedule a call: https://calendly.com/hudson-data/hello
- Send a brief: https://hudsondata.com/contact-us
- Email: info@hudsondata.com
- Careers: https://career.hudsondata.com/careers

## Locations

- US headquarters (customer-facing engineering, product, research bench)
- Gurugram, India (engineering and data science delivery)

Hybrid by design.