2024/07/10 - Amazon Bedrock - 3 updated api methods
Changes Add support for contextual grounding check for Guardrails for Amazon Bedrock.
{'contextualGroundingPolicyConfig': {'filtersConfig': [{'threshold': 'double',
'type': 'GROUNDING | '
'RELEVANCE'}]}}
Creates a guardrail to block topics and to implement safeguards for your generative AI applications.
You can configure the following policies in a guardrail to avoid undesirable and harmful content, filter out denied topics and words, and remove sensitive information for privacy protection.
Content filters - Adjust filter strengths to block input prompts or model responses containing harmful content.
Denied topics - Define a set of topics that are undesirable in the context of your application. These topics will be blocked if detected in user queries or model responses.
Word filters - Configure filters to block undesirable words, phrases, and profanity. Such words can include offensive terms, competitor names etc.
Sensitive information filters - Block or mask sensitive information such as personally identifiable information (PII) or custom regex in user inputs and model responses.
In addition to the above policies, you can also configure the messages to be returned to the user if a user input or model response is in violation of the policies defined in the guardrail.
For more information, see Guardrails for Amazon Bedrock in the Amazon Bedrock User Guide.
See also: AWS API Documentation
Request Syntax
client.create_guardrail(
name='string',
description='string',
topicPolicyConfig={
'topicsConfig': [
{
'name': 'string',
'definition': 'string',
'examples': [
'string',
],
'type': 'DENY'
},
]
},
contentPolicyConfig={
'filtersConfig': [
{
'type': 'SEXUAL'|'VIOLENCE'|'HATE'|'INSULTS'|'MISCONDUCT'|'PROMPT_ATTACK',
'inputStrength': 'NONE'|'LOW'|'MEDIUM'|'HIGH',
'outputStrength': 'NONE'|'LOW'|'MEDIUM'|'HIGH'
},
]
},
wordPolicyConfig={
'wordsConfig': [
{
'text': 'string'
},
],
'managedWordListsConfig': [
{
'type': 'PROFANITY'
},
]
},
sensitiveInformationPolicyConfig={
'piiEntitiesConfig': [
{
'type': 'ADDRESS'|'AGE'|'AWS_ACCESS_KEY'|'AWS_SECRET_KEY'|'CA_HEALTH_NUMBER'|'CA_SOCIAL_INSURANCE_NUMBER'|'CREDIT_DEBIT_CARD_CVV'|'CREDIT_DEBIT_CARD_EXPIRY'|'CREDIT_DEBIT_CARD_NUMBER'|'DRIVER_ID'|'EMAIL'|'INTERNATIONAL_BANK_ACCOUNT_NUMBER'|'IP_ADDRESS'|'LICENSE_PLATE'|'MAC_ADDRESS'|'NAME'|'PASSWORD'|'PHONE'|'PIN'|'SWIFT_CODE'|'UK_NATIONAL_HEALTH_SERVICE_NUMBER'|'UK_NATIONAL_INSURANCE_NUMBER'|'UK_UNIQUE_TAXPAYER_REFERENCE_NUMBER'|'URL'|'USERNAME'|'US_BANK_ACCOUNT_NUMBER'|'US_BANK_ROUTING_NUMBER'|'US_INDIVIDUAL_TAX_IDENTIFICATION_NUMBER'|'US_PASSPORT_NUMBER'|'US_SOCIAL_SECURITY_NUMBER'|'VEHICLE_IDENTIFICATION_NUMBER',
'action': 'BLOCK'|'ANONYMIZE'
},
],
'regexesConfig': [
{
'name': 'string',
'description': 'string',
'pattern': 'string',
'action': 'BLOCK'|'ANONYMIZE'
},
]
},
contextualGroundingPolicyConfig={
'filtersConfig': [
{
'type': 'GROUNDING'|'RELEVANCE',
'threshold': 123.0
},
]
},
blockedInputMessaging='string',
blockedOutputsMessaging='string',
kmsKeyId='string',
tags=[
{
'key': 'string',
'value': 'string'
},
],
clientRequestToken='string'
)
string
[REQUIRED]
The name to give the guardrail.
string
A description of the guardrail.
dict
The topic policies to configure for the guardrail.
topicsConfig (list) -- [REQUIRED]
A list of policies related to topics that the guardrail should deny.
(dict) --
Details about topics for the guardrail to identify and deny.
name (string) -- [REQUIRED]
The name of the topic to deny.
definition (string) -- [REQUIRED]
A definition of the topic to deny.
examples (list) --
A list of prompts, each of which is an example of a prompt that can be categorized as belonging to the topic.
(string) --
type (string) -- [REQUIRED]
Specifies to deny the topic.
dict
The content filter policies to configure for the guardrail.
filtersConfig (list) -- [REQUIRED]
Contains the type of the content filter and how strongly it should apply to prompts and model responses.
(dict) --
Contains filter strengths for harmful content. Guardrails support the following content filters to detect and filter harmful user inputs and FM-generated outputs.
Hate – Describes language or a statement that discriminates, criticizes, insults, denounces, or dehumanizes a person or group on the basis of an identity (such as race, ethnicity, gender, religion, sexual orientation, ability, and national origin).
Insults – Describes language or a statement that includes demeaning, humiliating, mocking, insulting, or belittling language. This type of language is also labeled as bullying.
Sexual – Describes language or a statement that indicates sexual interest, activity, or arousal using direct or indirect references to body parts, physical traits, or sex.
Violence – Describes language or a statement that includes glorification of or threats to inflict physical pain, hurt, or injury toward a person, group or thing.
Content filtering depends on the confidence classification of user inputs and FM responses across each of the four harmful categories. All input and output statements are classified into one of four confidence levels (NONE, LOW, MEDIUM, HIGH) for each harmful category. For example, if a statement is classified as Hate with HIGH confidence, the likelihood of the statement representing hateful content is high. A single statement can be classified across multiple categories with varying confidence levels. For example, a single statement can be classified as Hate with HIGH confidence, Insults with LOW confidence, Sexual with NONE confidence, and Violence with MEDIUM confidence.
For more information, see Guardrails content filters.
type (string) -- [REQUIRED]
The harmful category that the content filter is applied to.
inputStrength (string) -- [REQUIRED]
The strength of the content filter to apply to prompts. As you increase the filter strength, the likelihood of filtering harmful content increases and the probability of seeing harmful content in your application reduces.
outputStrength (string) -- [REQUIRED]
The strength of the content filter to apply to model responses. As you increase the filter strength, the likelihood of filtering harmful content increases and the probability of seeing harmful content in your application reduces.
dict
The word policy you configure for the guardrail.
wordsConfig (list) --
A list of words to configure for the guardrail.
(dict) --
A word to configure for the guardrail.
text (string) -- [REQUIRED]
Text of the word configured for the guardrail to block.
managedWordListsConfig (list) --
A list of managed words to configure for the guardrail.
(dict) --
The managed word list to configure for the guardrail.
type (string) -- [REQUIRED]
The managed word type to configure for the guardrail.
dict
The sensitive information policy to configure for the guardrail.
piiEntitiesConfig (list) --
A list of PII entities to configure to the guardrail.
(dict) --
The PII entity to configure for the guardrail.
type (string) -- [REQUIRED]
Configure guardrail type when the PII entity is detected.
The following PIIs are used to block or mask sensitive information:
General
ADDRESS A physical address, such as "100 Main Street, Anytown, USA" or "Suite #12, Building 123". An address can include information such as the street, building, location, city, state, country, county, zip code, precinct, and neighborhood.
AGE An individual's age, including the quantity and unit of time. For example, in the phrase "I am 40 years old," Guarrails recognizes "40 years" as an age.
NAME An individual's name. This entity type does not include titles, such as Dr., Mr., Mrs., or Miss. guardrails doesn't apply this entity type to names that are part of organizations or addresses. For example, guardrails recognizes the "John Doe Organization" as an organization, and it recognizes "Jane Doe Street" as an address.
EMAIL An email address, such as marymajor@email.com.
PHONE A phone number. This entity type also includes fax and pager numbers.
USERNAME A user name that identifies an account, such as a login name, screen name, nick name, or handle.
PASSWORD An alphanumeric string that is used as a password, such as "very20special#pass".
DRIVER_ID The number assigned to a driver's license, which is an official document permitting an individual to operate one or more motorized vehicles on a public road. A driver's license number consists of alphanumeric characters.
LICENSE_PLATE A license plate for a vehicle is issued by the state or country where the vehicle is registered. The format for passenger vehicles is typically five to eight digits, consisting of upper-case letters and numbers. The format varies depending on the location of the issuing state or country.
VEHICLE_IDENTIFICATION_NUMBER A Vehicle Identification Number (VIN) uniquely identifies a vehicle. VIN content and format are defined in the ISO 3779 specification. Each country has specific codes and formats for VINs.
Finance
REDIT_DEBIT_CARD_CVV A three-digit card verification code (CVV) that is present on VISA, MasterCard, and Discover credit and debit cards. For American Express credit or debit cards, the CVV is a four-digit numeric code.
CREDIT_DEBIT_CARD_EXPIRY The expiration date for a credit or debit card. This number is usually four digits long and is often formatted as month/year or MM/YY. Guardrails recognizes expiration dates such as 01/21, 01/2021, and Jan 2021.
CREDIT_DEBIT_CARD_NUMBER The number for a credit or debit card. These numbers can vary from 13 to 16 digits in length. However, Amazon Comprehend also recognizes credit or debit card numbers when only the last four digits are present.
PIN A four-digit personal identification number (PIN) with which you can access your bank account.
INTERNATIONAL_BANK_ACCOUNT_NUMBER An International Bank Account Number has specific formats in each country. For more information, see www.iban.com/structure.
SWIFT_CODE A SWIFT code is a standard format of Bank Identifier Code (BIC) used to specify a particular bank or branch. Banks use these codes for money transfers such as international wire transfers. SWIFT codes consist of eight or 11 characters. The 11-digit codes refer to specific branches, while eight-digit codes (or 11-digit codes ending in 'XXX') refer to the head or primary office.
IT
IP_ADDRESS An IPv4 address, such as 198.51.100.0.
MAC_ADDRESS A media access control (MAC) address is a unique identifier assigned to a network interface controller (NIC).
URL A web address, such as www.example.com.
AWS_ACCESS_KEY A unique identifier that's associated with a secret access key; you use the access key ID and secret access key to sign programmatic Amazon Web Services requests cryptographically.
AWS_SECRET_KEY A unique identifier that's associated with an access key. You use the access key ID and secret access key to sign programmatic Amazon Web Services requests cryptographically.
USA specific
US_BANK_ACCOUNT_NUMBER A US bank account number, which is typically 10 to 12 digits long.
US_BANK_ROUTING_NUMBER A US bank account routing number. These are typically nine digits long,
US_INDIVIDUAL_TAX_IDENTIFICATION_NUMBER A US Individual Taxpayer Identification Number (ITIN) is a nine-digit number that starts with a "9" and contain a "7" or "8" as the fourth digit. An ITIN can be formatted with a space or a dash after the third and forth digits.
US_PASSPORT_NUMBER A US passport number. Passport numbers range from six to nine alphanumeric characters.
US_SOCIAL_SECURITY_NUMBER A US Social Security Number (SSN) is a nine-digit number that is issued to US citizens, permanent residents, and temporary working residents.
Canada specific
CA_HEALTH_NUMBER A Canadian Health Service Number is a 10-digit unique identifier, required for individuals to access healthcare benefits.
CA_SOCIAL_INSURANCE_NUMBER A Canadian Social Insurance Number (SIN) is a nine-digit unique identifier, required for individuals to access government programs and benefits. The SIN is formatted as three groups of three digits, such as 123-456-789. A SIN can be validated through a simple check-digit process called the Luhn algorithm.
UK Specific
UK_NATIONAL_HEALTH_SERVICE_NUMBER A UK National Health Service Number is a 10-17 digit number, such as 485 777 3456. The current system formats the 10-digit number with spaces after the third and sixth digits. The final digit is an error-detecting checksum.
UK_NATIONAL_INSURANCE_NUMBER A UK National Insurance Number (NINO) provides individuals with access to National Insurance (social security) benefits. It is also used for some purposes in the UK tax system. The number is nine digits long and starts with two letters, followed by six numbers and one letter. A NINO can be formatted with a space or a dash after the two letters and after the second, forth, and sixth digits.
UK_UNIQUE_TAXPAYER_REFERENCE_NUMBER A UK Unique Taxpayer Reference (UTR) is a 10-digit number that identifies a taxpayer or a business.
Custom
Regex filter - You can use a regular expressions to define patterns for a guardrail to recognize and act upon such as serial number, booking ID etc..
action (string) -- [REQUIRED]
Configure guardrail action when the PII entity is detected.
regexesConfig (list) --
A list of regular expressions to configure to the guardrail.
(dict) --
The regular expression to configure for the guardrail.
name (string) -- [REQUIRED]
The name of the regular expression to configure for the guardrail.
description (string) --
The description of the regular expression to configure for the guardrail.
pattern (string) -- [REQUIRED]
The regular expression pattern to configure for the guardrail.
action (string) -- [REQUIRED]
The guardrail action to configure when matching regular expression is detected.
dict
The contextual grounding policy configuration used to create a guardrail.
filtersConfig (list) -- [REQUIRED]
The filter configuration details for the guardrails contextual grounding policy.
(dict) --
The filter configuration details for the guardrails contextual grounding filter.
type (string) -- [REQUIRED]
The filter details for the guardrails contextual grounding filter.
threshold (float) -- [REQUIRED]
The threshold details for the guardrails contextual grounding filter.
string
[REQUIRED]
The message to return when the guardrail blocks a prompt.
string
[REQUIRED]
The message to return when the guardrail blocks a model response.
string
The ARN of the KMS key that you use to encrypt the guardrail.
list
The tags that you want to attach to the guardrail.
(dict) --
Definition of the key/value pair for a tag.
key (string) -- [REQUIRED]
Key for the tag.
value (string) -- [REQUIRED]
Value for the tag.
string
A unique, case-sensitive identifier to ensure that the API request completes no more than once. If this token matches a previous request, Amazon Bedrock ignores the request, but does not return an error. For more information, see Ensuring idempotency in the Amazon S3 User Guide.
This field is autopopulated if not provided.
dict
Response Syntax
{
'guardrailId': 'string',
'guardrailArn': 'string',
'version': 'string',
'createdAt': datetime(2015, 1, 1)
}
Response Structure
(dict) --
guardrailId (string) --
The unique identifier of the guardrail that was created.
guardrailArn (string) --
The ARN of the guardrail.
version (string) --
The version of the guardrail that was created. This value will always be DRAFT.
createdAt (datetime) --
The time at which the guardrail was created.
{'contextualGroundingPolicy': {'filters': [{'threshold': 'double',
'type': 'GROUNDING | RELEVANCE'}]}}
Gets details about a guardrail. If you don't specify a version, the response returns details for the DRAFT version.
See also: AWS API Documentation
Request Syntax
client.get_guardrail(
guardrailIdentifier='string',
guardrailVersion='string'
)
string
[REQUIRED]
The unique identifier of the guardrail for which to get details. This can be an ID or the ARN.
string
The version of the guardrail for which to get details. If you don't specify a version, the response returns details for the DRAFT version.
dict
Response Syntax
{
'name': 'string',
'description': 'string',
'guardrailId': 'string',
'guardrailArn': 'string',
'version': 'string',
'status': 'CREATING'|'UPDATING'|'VERSIONING'|'READY'|'FAILED'|'DELETING',
'topicPolicy': {
'topics': [
{
'name': 'string',
'definition': 'string',
'examples': [
'string',
],
'type': 'DENY'
},
]
},
'contentPolicy': {
'filters': [
{
'type': 'SEXUAL'|'VIOLENCE'|'HATE'|'INSULTS'|'MISCONDUCT'|'PROMPT_ATTACK',
'inputStrength': 'NONE'|'LOW'|'MEDIUM'|'HIGH',
'outputStrength': 'NONE'|'LOW'|'MEDIUM'|'HIGH'
},
]
},
'wordPolicy': {
'words': [
{
'text': 'string'
},
],
'managedWordLists': [
{
'type': 'PROFANITY'
},
]
},
'sensitiveInformationPolicy': {
'piiEntities': [
{
'type': 'ADDRESS'|'AGE'|'AWS_ACCESS_KEY'|'AWS_SECRET_KEY'|'CA_HEALTH_NUMBER'|'CA_SOCIAL_INSURANCE_NUMBER'|'CREDIT_DEBIT_CARD_CVV'|'CREDIT_DEBIT_CARD_EXPIRY'|'CREDIT_DEBIT_CARD_NUMBER'|'DRIVER_ID'|'EMAIL'|'INTERNATIONAL_BANK_ACCOUNT_NUMBER'|'IP_ADDRESS'|'LICENSE_PLATE'|'MAC_ADDRESS'|'NAME'|'PASSWORD'|'PHONE'|'PIN'|'SWIFT_CODE'|'UK_NATIONAL_HEALTH_SERVICE_NUMBER'|'UK_NATIONAL_INSURANCE_NUMBER'|'UK_UNIQUE_TAXPAYER_REFERENCE_NUMBER'|'URL'|'USERNAME'|'US_BANK_ACCOUNT_NUMBER'|'US_BANK_ROUTING_NUMBER'|'US_INDIVIDUAL_TAX_IDENTIFICATION_NUMBER'|'US_PASSPORT_NUMBER'|'US_SOCIAL_SECURITY_NUMBER'|'VEHICLE_IDENTIFICATION_NUMBER',
'action': 'BLOCK'|'ANONYMIZE'
},
],
'regexes': [
{
'name': 'string',
'description': 'string',
'pattern': 'string',
'action': 'BLOCK'|'ANONYMIZE'
},
]
},
'contextualGroundingPolicy': {
'filters': [
{
'type': 'GROUNDING'|'RELEVANCE',
'threshold': 123.0
},
]
},
'createdAt': datetime(2015, 1, 1),
'updatedAt': datetime(2015, 1, 1),
'statusReasons': [
'string',
],
'failureRecommendations': [
'string',
],
'blockedInputMessaging': 'string',
'blockedOutputsMessaging': 'string',
'kmsKeyArn': 'string'
}
Response Structure
(dict) --
name (string) --
The name of the guardrail.
description (string) --
The description of the guardrail.
guardrailId (string) --
The unique identifier of the guardrail.
guardrailArn (string) --
The ARN of the guardrail.
version (string) --
The version of the guardrail.
status (string) --
The status of the guardrail.
topicPolicy (dict) --
The topic policy that was configured for the guardrail.
topics (list) --
A list of policies related to topics that the guardrail should deny.
(dict) --
Details about topics for the guardrail to identify and deny.
This data type is used in the following API operations:
name (string) --
The name of the topic to deny.
definition (string) --
A definition of the topic to deny.
examples (list) --
A list of prompts, each of which is an example of a prompt that can be categorized as belonging to the topic.
(string) --
type (string) --
Specifies to deny the topic.
contentPolicy (dict) --
The content policy that was configured for the guardrail.
filters (list) --
Contains the type of the content filter and how strongly it should apply to prompts and model responses.
(dict) --
Contains filter strengths for harmful content. Guardrails support the following content filters to detect and filter harmful user inputs and FM-generated outputs.
Hate – Describes language or a statement that discriminates, criticizes, insults, denounces, or dehumanizes a person or group on the basis of an identity (such as race, ethnicity, gender, religion, sexual orientation, ability, and national origin).
Insults – Describes language or a statement that includes demeaning, humiliating, mocking, insulting, or belittling language. This type of language is also labeled as bullying.
Sexual – Describes language or a statement that indicates sexual interest, activity, or arousal using direct or indirect references to body parts, physical traits, or sex.
Violence – Describes language or a statement that includes glorification of or threats to inflict physical pain, hurt, or injury toward a person, group or thing.
Content filtering depends on the confidence classification of user inputs and FM responses across each of the four harmful categories. All input and output statements are classified into one of four confidence levels (NONE, LOW, MEDIUM, HIGH) for each harmful category. For example, if a statement is classified as Hate with HIGH confidence, the likelihood of the statement representing hateful content is high. A single statement can be classified across multiple categories with varying confidence levels. For example, a single statement can be classified as Hate with HIGH confidence, Insults with LOW confidence, Sexual with NONE confidence, and Violence with MEDIUM confidence.
For more information, see Guardrails content filters.
This data type is used in the following API operations:
type (string) --
The harmful category that the content filter is applied to.
inputStrength (string) --
The strength of the content filter to apply to prompts. As you increase the filter strength, the likelihood of filtering harmful content increases and the probability of seeing harmful content in your application reduces.
outputStrength (string) --
The strength of the content filter to apply to model responses. As you increase the filter strength, the likelihood of filtering harmful content increases and the probability of seeing harmful content in your application reduces.
wordPolicy (dict) --
The word policy that was configured for the guardrail.
words (list) --
A list of words configured for the guardrail.
(dict) --
A word configured for the guardrail.
text (string) --
Text of the word configured for the guardrail to block.
managedWordLists (list) --
A list of managed words configured for the guardrail.
(dict) --
The managed word list that was configured for the guardrail. (This is a list of words that are pre-defined and managed by guardrails only.)
type (string) --
ManagedWords$type The managed word type that was configured for the guardrail. (For now, we only offer profanity word list)
sensitiveInformationPolicy (dict) --
The sensitive information policy that was configured for the guardrail.
piiEntities (list) --
The list of PII entities configured for the guardrail.
(dict) --
The PII entity configured for the guardrail.
type (string) --
The type of PII entity. For example, Social Security Number.
action (string) --
The configured guardrail action when PII entity is detected.
regexes (list) --
The list of regular expressions configured for the guardrail.
(dict) --
The regular expression configured for the guardrail.
name (string) --
The name of the regular expression for the guardrail.
description (string) --
The description of the regular expression for the guardrail.
pattern (string) --
The pattern of the regular expression configured for the guardrail.
action (string) --
The action taken when a match to the regular expression is detected.
contextualGroundingPolicy (dict) --
The contextual grounding policy used in the guardrail.
filters (list) --
The filter details for the guardrails contextual grounding policy.
(dict) --
The details for the guardrails contextual grounding filter.
type (string) --
The filter type details for the guardrails contextual grounding filter.
threshold (float) --
The threshold details for the guardrails contextual grounding filter.
createdAt (datetime) --
The date and time at which the guardrail was created.
updatedAt (datetime) --
The date and time at which the guardrail was updated.
statusReasons (list) --
Appears if the status is FAILED. A list of reasons for why the guardrail failed to be created, updated, versioned, or deleted.
(string) --
failureRecommendations (list) --
Appears if the status of the guardrail is FAILED. A list of recommendations to carry out before retrying the request.
(string) --
blockedInputMessaging (string) --
The message that the guardrail returns when it blocks a prompt.
blockedOutputsMessaging (string) --
The message that the guardrail returns when it blocks a model response.
kmsKeyArn (string) --
The ARN of the KMS key that encrypts the guardrail.
{'contextualGroundingPolicyConfig': {'filtersConfig': [{'threshold': 'double',
'type': 'GROUNDING | '
'RELEVANCE'}]}}
Updates a guardrail with the values you specify.
Specify a name and optional description.
Specify messages for when the guardrail successfully blocks a prompt or a model response in the blockedInputMessaging and blockedOutputsMessaging fields.
Specify topics for the guardrail to deny in the topicPolicyConfig object. Each GuardrailTopicConfig object in the topicsConfig list pertains to one topic.
Give a name and description so that the guardrail can properly identify the topic.
Specify DENY in the type field.
(Optional) Provide up to five prompts that you would categorize as belonging to the topic in the examples list.
Specify filter strengths for the harmful categories defined in Amazon Bedrock in the contentPolicyConfig object. Each GuardrailContentFilterConfig object in the filtersConfig list pertains to a harmful category. For more information, see Content filters. For more information about the fields in a content filter, see GuardrailContentFilterConfig.
Specify the category in the type field.
Specify the strength of the filter for prompts in the inputStrength field and for model responses in the strength field of the GuardrailContentFilterConfig.
(Optional) For security, include the ARN of a KMS key in the kmsKeyId field.
See also: AWS API Documentation
Request Syntax
client.update_guardrail(
guardrailIdentifier='string',
name='string',
description='string',
topicPolicyConfig={
'topicsConfig': [
{
'name': 'string',
'definition': 'string',
'examples': [
'string',
],
'type': 'DENY'
},
]
},
contentPolicyConfig={
'filtersConfig': [
{
'type': 'SEXUAL'|'VIOLENCE'|'HATE'|'INSULTS'|'MISCONDUCT'|'PROMPT_ATTACK',
'inputStrength': 'NONE'|'LOW'|'MEDIUM'|'HIGH',
'outputStrength': 'NONE'|'LOW'|'MEDIUM'|'HIGH'
},
]
},
wordPolicyConfig={
'wordsConfig': [
{
'text': 'string'
},
],
'managedWordListsConfig': [
{
'type': 'PROFANITY'
},
]
},
sensitiveInformationPolicyConfig={
'piiEntitiesConfig': [
{
'type': 'ADDRESS'|'AGE'|'AWS_ACCESS_KEY'|'AWS_SECRET_KEY'|'CA_HEALTH_NUMBER'|'CA_SOCIAL_INSURANCE_NUMBER'|'CREDIT_DEBIT_CARD_CVV'|'CREDIT_DEBIT_CARD_EXPIRY'|'CREDIT_DEBIT_CARD_NUMBER'|'DRIVER_ID'|'EMAIL'|'INTERNATIONAL_BANK_ACCOUNT_NUMBER'|'IP_ADDRESS'|'LICENSE_PLATE'|'MAC_ADDRESS'|'NAME'|'PASSWORD'|'PHONE'|'PIN'|'SWIFT_CODE'|'UK_NATIONAL_HEALTH_SERVICE_NUMBER'|'UK_NATIONAL_INSURANCE_NUMBER'|'UK_UNIQUE_TAXPAYER_REFERENCE_NUMBER'|'URL'|'USERNAME'|'US_BANK_ACCOUNT_NUMBER'|'US_BANK_ROUTING_NUMBER'|'US_INDIVIDUAL_TAX_IDENTIFICATION_NUMBER'|'US_PASSPORT_NUMBER'|'US_SOCIAL_SECURITY_NUMBER'|'VEHICLE_IDENTIFICATION_NUMBER',
'action': 'BLOCK'|'ANONYMIZE'
},
],
'regexesConfig': [
{
'name': 'string',
'description': 'string',
'pattern': 'string',
'action': 'BLOCK'|'ANONYMIZE'
},
]
},
contextualGroundingPolicyConfig={
'filtersConfig': [
{
'type': 'GROUNDING'|'RELEVANCE',
'threshold': 123.0
},
]
},
blockedInputMessaging='string',
blockedOutputsMessaging='string',
kmsKeyId='string'
)
string
[REQUIRED]
The unique identifier of the guardrail. This can be an ID or the ARN.
string
[REQUIRED]
A name for the guardrail.
string
A description of the guardrail.
dict
The topic policy to configure for the guardrail.
topicsConfig (list) -- [REQUIRED]
A list of policies related to topics that the guardrail should deny.
(dict) --
Details about topics for the guardrail to identify and deny.
name (string) -- [REQUIRED]
The name of the topic to deny.
definition (string) -- [REQUIRED]
A definition of the topic to deny.
examples (list) --
A list of prompts, each of which is an example of a prompt that can be categorized as belonging to the topic.
(string) --
type (string) -- [REQUIRED]
Specifies to deny the topic.
dict
The content policy to configure for the guardrail.
filtersConfig (list) -- [REQUIRED]
Contains the type of the content filter and how strongly it should apply to prompts and model responses.
(dict) --
Contains filter strengths for harmful content. Guardrails support the following content filters to detect and filter harmful user inputs and FM-generated outputs.
Hate – Describes language or a statement that discriminates, criticizes, insults, denounces, or dehumanizes a person or group on the basis of an identity (such as race, ethnicity, gender, religion, sexual orientation, ability, and national origin).
Insults – Describes language or a statement that includes demeaning, humiliating, mocking, insulting, or belittling language. This type of language is also labeled as bullying.
Sexual – Describes language or a statement that indicates sexual interest, activity, or arousal using direct or indirect references to body parts, physical traits, or sex.
Violence – Describes language or a statement that includes glorification of or threats to inflict physical pain, hurt, or injury toward a person, group or thing.
Content filtering depends on the confidence classification of user inputs and FM responses across each of the four harmful categories. All input and output statements are classified into one of four confidence levels (NONE, LOW, MEDIUM, HIGH) for each harmful category. For example, if a statement is classified as Hate with HIGH confidence, the likelihood of the statement representing hateful content is high. A single statement can be classified across multiple categories with varying confidence levels. For example, a single statement can be classified as Hate with HIGH confidence, Insults with LOW confidence, Sexual with NONE confidence, and Violence with MEDIUM confidence.
For more information, see Guardrails content filters.
type (string) -- [REQUIRED]
The harmful category that the content filter is applied to.
inputStrength (string) -- [REQUIRED]
The strength of the content filter to apply to prompts. As you increase the filter strength, the likelihood of filtering harmful content increases and the probability of seeing harmful content in your application reduces.
outputStrength (string) -- [REQUIRED]
The strength of the content filter to apply to model responses. As you increase the filter strength, the likelihood of filtering harmful content increases and the probability of seeing harmful content in your application reduces.
dict
The word policy to configure for the guardrail.
wordsConfig (list) --
A list of words to configure for the guardrail.
(dict) --
A word to configure for the guardrail.
text (string) -- [REQUIRED]
Text of the word configured for the guardrail to block.
managedWordListsConfig (list) --
A list of managed words to configure for the guardrail.
(dict) --
The managed word list to configure for the guardrail.
type (string) -- [REQUIRED]
The managed word type to configure for the guardrail.
dict
The sensitive information policy to configure for the guardrail.
piiEntitiesConfig (list) --
A list of PII entities to configure to the guardrail.
(dict) --
The PII entity to configure for the guardrail.
type (string) -- [REQUIRED]
Configure guardrail type when the PII entity is detected.
The following PIIs are used to block or mask sensitive information:
General
ADDRESS A physical address, such as "100 Main Street, Anytown, USA" or "Suite #12, Building 123". An address can include information such as the street, building, location, city, state, country, county, zip code, precinct, and neighborhood.
AGE An individual's age, including the quantity and unit of time. For example, in the phrase "I am 40 years old," Guarrails recognizes "40 years" as an age.
NAME An individual's name. This entity type does not include titles, such as Dr., Mr., Mrs., or Miss. guardrails doesn't apply this entity type to names that are part of organizations or addresses. For example, guardrails recognizes the "John Doe Organization" as an organization, and it recognizes "Jane Doe Street" as an address.
EMAIL An email address, such as marymajor@email.com.
PHONE A phone number. This entity type also includes fax and pager numbers.
USERNAME A user name that identifies an account, such as a login name, screen name, nick name, or handle.
PASSWORD An alphanumeric string that is used as a password, such as "very20special#pass".
DRIVER_ID The number assigned to a driver's license, which is an official document permitting an individual to operate one or more motorized vehicles on a public road. A driver's license number consists of alphanumeric characters.
LICENSE_PLATE A license plate for a vehicle is issued by the state or country where the vehicle is registered. The format for passenger vehicles is typically five to eight digits, consisting of upper-case letters and numbers. The format varies depending on the location of the issuing state or country.
VEHICLE_IDENTIFICATION_NUMBER A Vehicle Identification Number (VIN) uniquely identifies a vehicle. VIN content and format are defined in the ISO 3779 specification. Each country has specific codes and formats for VINs.
Finance
REDIT_DEBIT_CARD_CVV A three-digit card verification code (CVV) that is present on VISA, MasterCard, and Discover credit and debit cards. For American Express credit or debit cards, the CVV is a four-digit numeric code.
CREDIT_DEBIT_CARD_EXPIRY The expiration date for a credit or debit card. This number is usually four digits long and is often formatted as month/year or MM/YY. Guardrails recognizes expiration dates such as 01/21, 01/2021, and Jan 2021.
CREDIT_DEBIT_CARD_NUMBER The number for a credit or debit card. These numbers can vary from 13 to 16 digits in length. However, Amazon Comprehend also recognizes credit or debit card numbers when only the last four digits are present.
PIN A four-digit personal identification number (PIN) with which you can access your bank account.
INTERNATIONAL_BANK_ACCOUNT_NUMBER An International Bank Account Number has specific formats in each country. For more information, see www.iban.com/structure.
SWIFT_CODE A SWIFT code is a standard format of Bank Identifier Code (BIC) used to specify a particular bank or branch. Banks use these codes for money transfers such as international wire transfers. SWIFT codes consist of eight or 11 characters. The 11-digit codes refer to specific branches, while eight-digit codes (or 11-digit codes ending in 'XXX') refer to the head or primary office.
IT
IP_ADDRESS An IPv4 address, such as 198.51.100.0.
MAC_ADDRESS A media access control (MAC) address is a unique identifier assigned to a network interface controller (NIC).
URL A web address, such as www.example.com.
AWS_ACCESS_KEY A unique identifier that's associated with a secret access key; you use the access key ID and secret access key to sign programmatic Amazon Web Services requests cryptographically.
AWS_SECRET_KEY A unique identifier that's associated with an access key. You use the access key ID and secret access key to sign programmatic Amazon Web Services requests cryptographically.
USA specific
US_BANK_ACCOUNT_NUMBER A US bank account number, which is typically 10 to 12 digits long.
US_BANK_ROUTING_NUMBER A US bank account routing number. These are typically nine digits long,
US_INDIVIDUAL_TAX_IDENTIFICATION_NUMBER A US Individual Taxpayer Identification Number (ITIN) is a nine-digit number that starts with a "9" and contain a "7" or "8" as the fourth digit. An ITIN can be formatted with a space or a dash after the third and forth digits.
US_PASSPORT_NUMBER A US passport number. Passport numbers range from six to nine alphanumeric characters.
US_SOCIAL_SECURITY_NUMBER A US Social Security Number (SSN) is a nine-digit number that is issued to US citizens, permanent residents, and temporary working residents.
Canada specific
CA_HEALTH_NUMBER A Canadian Health Service Number is a 10-digit unique identifier, required for individuals to access healthcare benefits.
CA_SOCIAL_INSURANCE_NUMBER A Canadian Social Insurance Number (SIN) is a nine-digit unique identifier, required for individuals to access government programs and benefits. The SIN is formatted as three groups of three digits, such as 123-456-789. A SIN can be validated through a simple check-digit process called the Luhn algorithm.
UK Specific
UK_NATIONAL_HEALTH_SERVICE_NUMBER A UK National Health Service Number is a 10-17 digit number, such as 485 777 3456. The current system formats the 10-digit number with spaces after the third and sixth digits. The final digit is an error-detecting checksum.
UK_NATIONAL_INSURANCE_NUMBER A UK National Insurance Number (NINO) provides individuals with access to National Insurance (social security) benefits. It is also used for some purposes in the UK tax system. The number is nine digits long and starts with two letters, followed by six numbers and one letter. A NINO can be formatted with a space or a dash after the two letters and after the second, forth, and sixth digits.
UK_UNIQUE_TAXPAYER_REFERENCE_NUMBER A UK Unique Taxpayer Reference (UTR) is a 10-digit number that identifies a taxpayer or a business.
Custom
Regex filter - You can use a regular expressions to define patterns for a guardrail to recognize and act upon such as serial number, booking ID etc..
action (string) -- [REQUIRED]
Configure guardrail action when the PII entity is detected.
regexesConfig (list) --
A list of regular expressions to configure to the guardrail.
(dict) --
The regular expression to configure for the guardrail.
name (string) -- [REQUIRED]
The name of the regular expression to configure for the guardrail.
description (string) --
The description of the regular expression to configure for the guardrail.
pattern (string) -- [REQUIRED]
The regular expression pattern to configure for the guardrail.
action (string) -- [REQUIRED]
The guardrail action to configure when matching regular expression is detected.
dict
The contextual grounding policy configuration used to update a guardrail.
filtersConfig (list) -- [REQUIRED]
The filter configuration details for the guardrails contextual grounding policy.
(dict) --
The filter configuration details for the guardrails contextual grounding filter.
type (string) -- [REQUIRED]
The filter details for the guardrails contextual grounding filter.
threshold (float) -- [REQUIRED]
The threshold details for the guardrails contextual grounding filter.
string
[REQUIRED]
The message to return when the guardrail blocks a prompt.
string
[REQUIRED]
The message to return when the guardrail blocks a model response.
string
The ARN of the KMS key with which to encrypt the guardrail.
dict
Response Syntax
{
'guardrailId': 'string',
'guardrailArn': 'string',
'version': 'string',
'updatedAt': datetime(2015, 1, 1)
}
Response Structure
(dict) --
guardrailId (string) --
The unique identifier of the guardrail
guardrailArn (string) --
The ARN of the guardrail.
version (string) --
The version of the guardrail.
updatedAt (datetime) --
The date and time at which the guardrail was updated.