Entity types

An entity type is a label assigned to a detected piece of sensitive information, such as EMAIL_ADDRESS or PERSON. AI Smart Redact ships with 36 built-in entity types: 32 pattern-based labels detected by regex pattern recognizers, and 4 semantic labels detected by the semantic model. You can also define your own entity types per detection request.

Pattern-based labels

The following 32 labels have built-in pattern recognizers. Each label may match more than one format. For example, DATE covers numeric and verbal date formats across more than one language.

Label	What it matches
ALPHANUMERIC_CODE	Mixed letter-digit codes
BARCODE	Product barcodes with checksum validation
BIC_SWIFT	Bank Identifier Codes
CREDIT_CARD	Credit card numbers with Luhn validation
CURRENCY_CODE	ISO 4217 three-letter currency codes
DATE	Calendar dates in common international formats
DATETIME	Combined date-time strings
DECIMAL_NUMBER	Decimal numbers
DOMAIN_NAME	Internet domain names
DURATION	Time durations
EMAIL_ADDRESS	Email addresses
FILE_PATH	Windows and Unix file paths
GPS_COORDINATE	Geographic coordinates
HASHTAG	Social media hashtags
HTTP_COOKIE	HTTP cookie strings
IBAN	International Bank Account Numbers
INTEGER_NUMBER	Integers
IP_ADDRESS	IPv4 and IPv6 addresses
ISIN	International Securities Identification Numbers
LEI	Legal Entity Identifiers
MAC_ADDRESS	Hardware MAC addresses
MENTION	Social media @mentions
MONEY	Monetary amounts
NUMERIC_ID	Digit-only identifiers
PERCENTAGE	Percentage values
PHONE_NUMBER	Phone numbers in common international formats
SCIENTIFIC_NUMBER	Scientific notation numbers
TIME	Times in common formats
UNIQUE_IDENTIFIER	UUID and GUID strings
URL	HTTP, HTTPS, and FTP URLs
VAT_NUMBER	VAT numbers
VIN	Vehicle Identification Numbers

For details on how pattern recognizers work and how to add custom ones, refer to Pattern recognizers.

Semantic labels

Out of the box, the semantic model detects four entity types:

Label	What it matches
PERSON	Person names
ORGANISATION	Organization and company names
PHYSICAL_ADDRESS	Street addresses and locations
USERNAME	Usernames and handles

The built-in default mapping is a starting point, not a fixed set. You can extend or replace it to detect any entity type the semantic model can recognize. Refer to Semantic recognizer.

Confidence scores

Every detected entity has a confidence score between 0 and 1. The score reflects how unambiguous the match is. Entities with structurally unique formats and supporting validators score higher; broader patterns that can match many non-PII strings score lower.

Range	Tier	Description
0.90–1.00	Highest	Structurally unique, almost no false positives
0.80–0.89	High	Distinctive structure, often with checksum validation
0.70–0.79	Moderate	Recognizable format with some ambiguity
0.60–0.69	Lower	Formats that overlap with common text
0.50–0.59	Low	Broad patterns with higher ambiguity
Under 0.50	Sub-threshold	Filtered out at the built-in default threshold. Refer to Sub-threshold patterns.

The built-in default scoreThreshold is 0.5. Entities scoring under the threshold are removed from the results. Raise the threshold to favor precision; lower it to favor recall.

Sub-threshold patterns

Some patterns are deliberately configured to score under the built-in default threshold to prevent false positives from highly ambiguous matches. For example, a bare four-digit number could be military time, a year, a PIN, or an arbitrary identifier; the military-time pattern scores under 0.5 and surfaces only when context boosting raises the score over the threshold.

Custom entity types

You can extend the set of detected entity types in three ways:

• Add custom regex patterns. Refer to Pattern recognizers.
• Add a denylist of known sensitive terms. Refer to Keyword recognizers.
• Map more semantic-model outputs to new labels. Refer to Semantic recognizer.