Skip to main content

Data Privacy Platform

Make your data safe for use with AI and LLMs

Granica Screen is a data privacy platform that protects natural language processing data & models – from training and fine-tuning to inference. It discovers sensitive information in cloud data lake files and input prompts with state-of-the-art accuracy to ensure safe use with both in-house and external AI services.

The Data Privacy Problem

It's hard to make data safe for use with AI, and the stakes keep getting higher

breach

LLMs Can Easily Leak PII

If AI teams train, fine tune, or prompt LLMs with PII and sensitive information it not only introduces bias that degrades AI results, it creates high risk of data leakage, i.e. breach, at inference time. 

decrease

Today's Discovery Isn't Effective

Preserving the privacy of training and prompt data has proven to be a persistent challenge. The two biggest issues? Low detection accuracy and poor cost-effectiveness at the scale AI requires. 

Introducing Granica Screen

SOTA Accuracy and Cost-Effectiveness at the Scale of AI

Granica Screen lets you discover all your sensitive info and PII with state-of-the-art accuracy, so nothing leaks out. Screen is also highly compute-efficient, so you can unlock up to 10X more data at the same cloud infrastructure cost as traditional solutions.

Use Cases for Data Privacy in GenAI

LLM privacy3
chatting

Protect LLM Prompts

Discover and mask sensitive information in end-user and application generated prompts, whether for use with in-house or 3rd-party LLMs. Granica Screen delivers:

  • State-of-the-art discovery accuracy
  • Safe, masked prompt inputs and de-masked outputs for contextual, natural interactions
  • Real-time, low-latency response times with auto-scaling to meet high-throughput scenarios

data-encryption

Protect LLM Training Data Stores

Discover and mask sensitive information in training data to ensure it doesn’t accidentally leak at inference time. Granica Screen delivers:

  • 100+ languages, 50+ named entities supported
  • Structured/unstructured text/NLP support
  • 5-10X lower compute costs for discovery at scale
training data discovery and masking3
synthetic data
pretend

Improve AI with Synthetic Data

Replace sensitive information with synthetic - realistic but fake - data to improve accuracy and privacy when training, fine-tuning, and prompting LLMs.

Cost-Effectively Preserve AI Privacy at Scale

The Granica Screen data privacy platform delivers 5-10x higher compute efficiency, which lowers infrastructure cost per byte scanned vs. traditional approaches and enables cost-effective scanning of broad data sets.

  •   Unlock 5-10x more data to power AI/ML.
  •   Safely train on fresh data as it lands.
  •   Protect large-scale data lakes.
cost-efficiency

Granica Screen in Action

With Granica Screen, you can confidently mitigate privacy risks while building better AI, faster.

Learn how Granica Screen can help you make data safe for AI.

whitepaper

Read Our Latest White Paper "Building Trust, Impact and Efficiency into Traditional and Generative AI"

Data types and classifiers supported

NLP/Text, Clickstream, Logs, Tabular, and more

Granica supports a wide range of AI/ML/analytics data types and classifiers (e.g., phone number, SSN, VIN, etc.). Bring us your unique requirements, and we can customize it for your use case.

text_squar_small2

NLP/Text

use-case-clickstream

Clickstream

use case logs

Logs

use-case-tabular

Tabular

Related Articles

Top Data Privacy Tools for 2024

AI Data Privacy Challenges and Best Practices

Unveiling the New Era of Data Privacy: A Guide for CIOs and CISOs

Frequently Asked Questions:

Does Granica Screen modify the private source files?

No. Granica Screen is typically granted read-only access to private files. It reads and transforms sensitive information such as PII in those files using various de-identification techniques and then stores safe-for-use copies in a separate target bucket.

How does Granica Screen compare to alternatives?

Granica Screen provides state-of-the-art named entity recognition (NER) accuracy for 50+ entities and global support for 100+ languages. High accuracy is the critical foundation of real data privacy, enabling truly safe and compliant use in ML and generative AI.

Screen is also is highly compute-efficient, lowering the cost to side-scan data by 5-10X, thus increasing the volume of data you can unlock for training by 5-10X at comparable costs. Finally, Granica Screen continuously monitors your data lake to detect and protect sensitive information and PII in new files immediately after they land.

Can I use Granica Screen and Granica Crunch together on the same data?

Yes, both products are built on the Granica platform and are fully compatible with one another. You can maximize your benefits by using them together. For example, a common pattern is to first use Granica Screen to generate safe-for-use file copies in a target bucket. Then, using Granica Crunch on that target bucket minimizes the cost of storing and accessing those copies.

Train, fine-tune, and prompt safely.

Unlock even more data for AI/ML teams to safely improve model performance, whether for private LLMs and generative AI or traditional AI and machine learning. The Granica Screen data privacy platform reduces breach and compliance risks while simultaneously improving the efficiency of downstream AI workflows.