PrismSek99.2%
detection precision
Understanding
Data Classification
AI-native labels that read context, so a nine-digit number is a routing code in one file and a Social Security number in another.
The problem
Regex-era classification drowns teams in false positives and misses anything that does not match a pattern. Classification that reads context is the difference between a label you can automate against and a label you have to re-review by hand.
How it works
What it covers
Data types
PIIPHIPCISecrets & tokensSource codeFinancial recordsIntellectual propertyCustom types via examples
Detection
- Fine-tuned transformer classifiers per data family
- Contextual disambiguation (test data vs. production data)
- Confidence scoring exposed on every detection
- Custom classifiers trained from a handful of your own examples
Actions
- Real-time labels consumable by DLP policies and external tools
- Native label sync with Microsoft Purview sensitivity labels
- Bulk backfill classification for existing stores
- Review queues only for detections below your confidence threshold
AI coverage
- Classifies prompts and completions in flight
- Labels chunks entering RAG pipelines before retrieval exposure
- Distinguishes synthetic/test data from production data in training sets
Deployment
SaaS multi-tenantDedicated single-tenantPrivate-cloud data plane
Regulations
GDPRHIPAAPCI DSSIndia DPDPCCPAGLBA
~180ms
p95 inference latency
70%
fewer false positives vs. pattern-based tools
Common questions
Can we define our own data types?
Yes. Custom classifiers are trained from labeled examples you provide in the console, typically a few dozen, and can be tested against historical data before going live.
What happens to low-confidence detections?
You set the threshold. Above it, policies act automatically; below it, detections route to a review queue instead of firing actions.
Does classification content leave our boundary?
Not with the private data plane. Inference runs where the data lives and only labels and metadata leave.
Solutions this powers
See Data Classification on your data.
Connect one environment in a guided session and review real findings with a security engineer.