Private Learning

Custom Machine Learning for
Fraud Protection

Detect fraud and abuse in real time with custom models tuned to your risk classes.
Privacy-preserving protection with Zero PII.

Get a Demo

How It Works

Powerful Joint Learning

Merge your pre-blinded data with our models and risk classes to enable joint learning, with precise detection and deterrence of novel threats.

Custom Risk Definitions

Opt for standardized risk classes or define your own in minutes. Our models  adapt to your unique needs, addressing even the most challenging edge cases.

Rule-to-Learn

Instantly tag a new actor while simultaneously retraining models to detect similar activity, thwarting even sophisticated actors within seconds.

A new way to stop online fraud and abuse, whether human or automated.

Contact Sales

Learn Globally, Predict Locally

Easily find and deter specific classes of risk behavior through powerful, real-time joint learning. Deploy scalable security machine learning at a pace you never thought possible.

Deploy in Minutes, No Labels Needed

With Label-Optional Learning, a few clicks are all you need to adjust real-time behavior and generalize predictions to find similar high-risk actors.

Combine Your Detections with Our Risk Intelligence

Seamlessly combine your pre-blinded data with our global models for sharper risk scoring, with provable privacy at every step. Our novel joint learning can outperform models trained solely on your own data, finding threats your existing strategies may overlook entirely.

Strengthen Your Downstream Fraud and Risk Analysis

Seamlessly complement your various solutions. As part of a fully integrated risk platform, Private Learning includes real-time and offline inference, rich analytics, rules, orchestration, and more to address your risk management needs.

Why Enterprise Teams Choose Private Learning

Detect and Deter Human and Automated Threats

Integrated AI/ML-driven detection, deterrence, insights, and remediation with rule-based control across the full user lifecycle.

Privacy-First, Globally Compliant

Private Learning is designed to comply with evolving privacy laws worldwide, using privacy-preserving machine learning to identify and prevent fraud without ever sharing sensitive customer data.

Industry-Leading Accuracy on Day 1

Unlike legacy approaches that take months to roll out, Private Learning bootstraps to rapidly produce useful results through our novel, label-free learning strategies.

With state-of-the-art automated safety mechanisms like drift detection, automated label quality control, rule-based score shaping, and more, you can rely on Private Learning from day 1.

Comprehensive and Familiar SDKs

Use the hCaptcha SDKs you already know. Just add a few new parameters to get custom class predictions, with total control over what pre-blinded data is sent and processed.

Scalable Backend Prediction APIs

Direct real-time and batch inference from your backend systems, no SDK session required. Protect API calls, machine-to-machine connections, in-game actions, and much more.

Private Learning Enhances hCaptcha Enterprise

A Fully Integrated Protection Suite

Fully Customized Machine Learning Models
Whether human or automated, each fraud scenario gets its own model.
Rule-to-Learn: Immediate Adjustment of Model Behavior
Set your own rules to express business needs and update models.
Label-Optional Learning
Our novel approach lets you start scoring in minutes, not weeks. No reporting required.
Standardized Risk Classes or Custom Risk Definitions
Predict risk according to your classes, or bootstrap from ours.
Real-Time Joint Learning for Higher Accuracy
Exceed the accuracy of models you could train yourself with our unique approach.
Anomaly Detection & Alerts
Stay on top of suspicious activity.
Legacy browser support
Compatible all the way back to Internet Explorer 8.
Support for non-JavaScript clients.
Powerful Dashboards with Rich Interpretability Features
Easily review traffic trends and visualize what’s driving detections.
Supports Fully Blinded Data
Private Learning has been optimized to work just as well with fully blinded data.
GDPR, CCPA, and LGPD-Compliant, Zero PII Operation
Designed for privacy from day one.
Ready to see what’s possible? Get a demo today.
Contact Sales

Frequently asked questions

What is machine learning fraud protection?

-
+
Machine learning fraud protection uses models to evaluate behavioral, session, and transaction signals and score risk in real time. It can detect patterns beyond explicit rules and adapt as attack behavior changes. hCaptcha Private Learning adds customer-defined risk classes and joint learning with pre-blinded data, enabling fraud and abuse detection without requiring personally identifiable information (PII).

How does hCaptcha Private Learning provide machine learning fraud protection without PII?

+
-
Customers control what data they send and can pre-blind fields before the data reaches hCaptcha. Private Learning combines that pre-blinded data with hCaptcha models and risk classes through joint learning to produce customer-specific risk predictions. This limits data exposure while preserving the signals needed to detect fraud and abuse.

How does machine learning fraud protection differ from rule-based fraud detection?

+
-
Rule-based fraud detection applies explicit conditions and thresholds, making it useful for known attack patterns and business policies. Machine learning evaluates relationships across many signals and can identify emerging patterns that no individual rule describes. Private Learning combines both approaches: Rule-to-Learn lets teams use business-defined rules to adjust model behavior.

How does Private Learning support GDPR, CCPA, LGPD, PIPL, and HIPAA requirements?

+
-
hCaptcha complies with GDPR, CCPA, LGPD, and PIPL. Private Learning supports fully pre-blinded, Zero PII deployments, which can help organizations subject to HIPAA limit data exposure and support their compliance obligations. hCaptcha maintains ISO/IEC 27001 and ISO/IEC 27701 certifications, a current SOC 2 Type II certification, PCI DSS 4.0 Level 1 Service Provider compliance, and certification under the EU-U.S., UK-U.S., and Swiss-U.S. Data Privacy Frameworks.

How long does it take to deploy machine learning fraud protection with Private Learning?

+
-
Private Learning can begin scoring risk in minutes. Label-Optional Learning does not require teams to hand-label a historical fraud dataset before receiving useful scores. Teams can choose standardized risk classes or define custom classes, then integrate through hCaptcha SDKs or backend prediction APIs. Rule-to-Learn lets teams adjust model behavior with business-defined rules after deployment.

What kinds of fraud does Private Learning detect?

+
-
Private Learning can model standardized or custom risk classes for account takeover, credential stuffing, card testing, transaction fraud, incentive abuse, in-game abuse, giveaway abuse, and coordinated automated activity. Teams can also define business-specific classes for edge cases. Each fraud scenario receives a tailored model within the same hCaptcha Enterprise platform.

How does Private Learning fit into an existing fraud stack?

+
-
Private Learning adds privacy-preserving machine learning to an existing fraud stack. It runs within hCaptcha Enterprise alongside Bot Detection, Account Defense, Fraud Protection, MFA, and User Journeys. Real-time and offline inference are available through familiar hCaptcha SDKs and scalable backend prediction APIs, allowing teams to use risk predictions in existing rules, orchestration, analytics, and investigation workflows.