AI Model

Proguard

The AI model that powers enterprise-grade security

Purpose-built for detecting sensitive data and security threats in AI interactions. Proguard delivers industry-leading accuracy with minimal latency, ensuring your AI systems remain secure without compromising performance.

Request Demo View Documentation
99.7%
Accuracy
<5ms
Latency
100+
Entity Types
50+
Languages

Advanced Capabilities

Proguard combines state-of-the-art machine learning with purpose-built security features.

PII Detection Engine

Identify 100+ types of personally identifiable information including names, emails, phone numbers, SSNs, credit cards, and more across 50+ languages.

99.7% accuracy

Threat Classification

Real-time detection of prompt injection attacks, jailbreak attempts, data exfiltration, and other malicious inputs targeting your AI systems.

<5ms latency

Context-Aware Analysis

Advanced semantic understanding reduces false positives while maintaining high recall rates. Understands context to make intelligent decisions.

95% fewer false positives

Multi-Language Support

Native support for 50+ languages with locale-specific entity recognition and cultural context awareness for global deployments.

50+ languages

Custom Entity Training

Train Proguard to recognize your organization-specific sensitive data patterns, product names, and proprietary information.

Unlimited custom entities

Continuous Learning

Model updates deployed automatically to stay ahead of emerging threats and new attack vectors without service interruption.

Weekly updates

Proguard

v3.2.1 Active

Safety & Governance Model

AI Safety PII Detection Content Moderation Guardrails Enterprise
<5ms
P99 Latency
700+req/s
Throughput
1.2GB
Memory
50+
Languages

Benchmark Results

Benchmark Metric Score
MLCommons Safety
Safety
 Detection Rate 99.85%
↳ Accuracy 99.85%
XSTest
Safety Refusal
 Unsafe Refusal Rate 87%
↳ Safe Compliance Rate 42%
↳ XSTest Score 64.5%
Benchmark Scores

Capabilities

PII Detection
Identify 100+ types of personally identifiable information across multiple languages and formats.
EMAIL PHONE SSN CREDIT_CARD +2
Threat Classification
Detect prompt injection, jailbreak attempts, and malicious inputs in real-time.
JAILBREAK INJECTION HARMFUL_CONTENT POLICY_VIOLATION
Context Analysis
Understand semantic context to reduce false positives while maintaining high recall.
INTENT SENTIMENT TOPIC RISK_LEVEL
Multi-language Support
Native support for 50+ languages with locale-specific entity recognition.
EN ES FR DE +3

Capability Profile

Model Information

Architecture
Transformer-based classifier ensemble
Parameters
~500M
License
Enterprise
Last Updated
2026-01-31

Live Detection Sample

Input:
"Contact john.doe@email.com at 555-123-4567"
Detected:
EMAIL PHONE
Action:
Anonymized → Forwarded

Intended Use

  • Enterprise AI gateway protection
  • PII/PHI detection and anonymization
  • Regulatory compliance (GDPR, HIPAA)
  • Real-time threat detection
Explore Proguard

Benchmark results from internal evaluation on MLCommons and XSTest datasets. View methodology

Common Use Cases

Proguard protects AI interactions across diverse enterprise applications.

Enterprise Chatbots

Protect customer-facing AI assistants from leaking sensitive business data or customer PII.

Code Generation

Prevent developers from accidentally exposing API keys, credentials, and proprietary code.

Document Analysis

Safely analyze contracts, reports, and documents while protecting confidential information.

Healthcare AI

HIPAA-compliant AI interactions that protect patient data and PHI at all times.

Ready to Secure Your AI?

See Proguard in action and learn how it can protect your AI infrastructure.

Request Demo Contact Sales