AI Guard: Classification Profiles

Classification profiles define which classifiers are used when AI Guard analyzes text. OneTrust provides over 300 system classifiers covering PII patterns, credentials, financial data, health information, and more.

How Profiles Work

When a classification request arrives, AI Guard loads the specified profile (or the default) and runs the associated classifiers against the submitted text. Each classifier that matches produces a result with:

The matched text
Position offsets (start and end)
Confidence score
Classifier name

Default Profile

Every AI Guard deployment includes a default classification profile that provides broad PII detection. Use it by specifying ClassifierDescriptionDefault() in the SDK:

from ai_guard.api import ClassifierDescriptionDefault

request = ClassificationRequest(
    context={"actor": "user"},
    classifier_description=ClassifierDescriptionDefault(),
    text="Your text here",
)

Changing the Default Profile

The default classification profile is defined within the OneTrust admin console and applies to a specific customer tenant. While a OneTrust administrator can change the default profile, it is recommended to reference specific profiles by name or UUID instead.

Using Specific Profiles

Reference a specific profile by UUID and version:

from ai_guard.api import ClassifierDescriptionProfile

request = ClassificationRequest(
    context={"actor": "user"},
    classifier_description=ClassifierDescriptionProfile(
        uuid="7dbf380f-0af8-4276-acb0-85413db2dbff",
        version=1,
    ),
    text="Your text here",
)

Profile UUIDs and versions are available in the OneTrust admin console under the Data Discovery classification settings.

Profile Types

Source	Description	Supported
System Profiles	Pre-built profiles provided by OneTrust covering 300+ patterns	Yes
Custom Profiles	Self-created profiles defined by your organization	Not currently supported

⚠️
Custom Profile Limitation
Custom classification profiles (self-created) are not currently supported for AI Guard operations. Only OneTrust system-defined profiles can be used.

Common Classifier Categories

AI Guard's 300+ system classifiers cover patterns including:

Category	Examples
Personal Identifiers	Social Security Numbers, National IDs, Passport Numbers
Contact Information	Phone Numbers, Email Addresses, Physical Addresses
Financial Data	Credit Card Numbers, Bank Account Numbers, IBAN
Healthcare	Medical Record Numbers, Health Plan IDs
Credentials	API Keys, Passwords, Secret Tokens
Location Data	IP Addresses, GPS Coordinates, Postal Codes

Confidence Threshold

The service has a configurable minimum confidence threshold (classification.min-allowed-likelihood). Only matches at or above this threshold are returned.

Likelihood	Description
`VERY_LIKELY`	Highest confidence — very few false positives
`LIKELY`	Default threshold — good balance of precision and recall
`POSSIBLE`	Lower confidence — more matches, more false positives
`UNLIKELY`	Lowest confidence — catches nearly everything

The default is LIKELY, which provides a good balance for most use cases.

Profile Caching

AI Guard caches loaded classification profiles and classifiers for fast response times. Profile updates in the OneTrust admin console are reflected after the cache is refreshed.

What's Next?

Classifier Descriptions — SDK options for specifying classifiers
Classify Text — Use profiles in classification requests
Service Configuration — Configure classification settings