AI Guard API Reference

Base URL

ai-guard.<csp>.<region>.pangea.cloud

post/v1/text/guard

curl -sSLX POST 'https://ai-guard.aws.us.pangea.cloud/v1/text/guard' \
-H 'Authorization: Bearer <your_token>' \
-H 'Content-Type: application/json' \
-d '{}'

Response

Raw Output

Guard LLM input and output text

POST

https://ai-guard.aws.us.pangea.cloud/v1/text/guard

Detect, remove, or block malicious content and intent in LLM inputs and outputs to prevent model manipulation and data leakage.

required parameters

messages

array<object>

Structured messages data to be scanned by AI Guard for PII, sensitive data, malicious content, and other data types defined by the configuration. Supports processing up to 20 KiB of JSON text using Pangea message format.

Click 'Save' to use the value

optional parameters

text

string

Text to be scanned by AI Guard for PII, sensitive data, malicious content, and other data types defined by the configuration. Supports processing up to 20 KiB of text.

recipe

string

(default: "pangea_prompt_guard")

Recipe key of a configuration of data types and settings defined in the Pangea User Console. It specifies the rules that are to be applied to the text, such as defang malicious URLs.

recipe

Recipe key of a configuration of data types and settings defined in the Pangea User Console. It specifies the rules that are to be applied to the text, such as defang malicious URLs.

debug

boolean

(default: false)

Setting this value to true will provide a detailed analysis of the text data

debug

Setting this value to true will provide a detailed analysis of the text data

overrides

object

ignore_recipe

boolean

(default: false)

Bypass existing Recipe content and create an on-the-fly Recipe.

ignore_recipe

Bypass existing Recipe content and create an on-the-fly Recipe.

code_detection

object

disabled

boolean

disabled

action

string

action

threshold

number

minimum: 0
maximum: 1
multipleOf: 0.01

threshold

minimum: 0
maximum: 1
multipleOf: 0.01

competitors

object

disabled

boolean

disabled

action

string

action

gibberish

object

disabled

boolean

disabled

action

string

action

language_detection

object

disabled

boolean

disabled

action

string

(default: "")

action

languages

array<string>

Click 'Save' to use the value

threshold

number

minimum: 0
maximum: 1
multipleOf: 0.01

threshold

minimum: 0
maximum: 1
multipleOf: 0.01

malicious_entity

object

disabled

boolean

disabled

domain

string

domain

ip_address

string

ip_address

url

string

url

pii_entity

object

disabled

boolean

disabled

au_abn

string

au_abn

au_acn

string

au_acn

au_medicare

string

au_medicare

au_tfn

string

au_tfn

credit_card

string

credit_card

crypto

string

crypto

date_time

string

date_time

email_address

string

email_address

fin/nric

string

fin/nric

iban_code

string

iban_code

ip_address

string

ip_address

location

string

location

medical_license

string

medical_license

money

string

money

nif

string

nif

nrp

string

nrp

person

string

person

phone_number

string

phone_number

uk_nhs

string

uk_nhs

url

string

url

us_bank_number

string

us_bank_number

us_drivers_license

string

us_drivers_license

us_itin

string

us_itin

us_passport

string

us_passport

us_ssn

string

us_ssn

prompt_injection

object

disabled

boolean

disabled

action

string

action

roleplay

object

disabled

boolean

disabled

action

string

action

secrets_detection

object

disabled

boolean

disabled

amazon_aws_access_key_id

string

amazon_aws_access_key_id

amazon_aws_secret_access_key

string

amazon_aws_secret_access_key

amazon_mws_auth_token

string

amazon_mws_auth_token

basic_auth

string

basic_auth

facebook_access_token

string

facebook_access_token

github_access_token

string

github_access_token

google_api_key

string

google_api_key

google_cloud_platform_api_key

string

google_cloud_platform_api_key

google_cloud_platform_service_account

string

google_cloud_platform_service_account

google_drive_api_key

string

google_drive_api_key

google_gmail_api_key

string

google_gmail_api_key

jwt_token

string

jwt_token

mailchimp_api_key

string

mailchimp_api_key

mailgun_api_key

string

mailgun_api_key

pangea_token

string

pangea_token

pgp_private_key_block

string

pgp_private_key_block

picatic_api_key

string

picatic_api_key

rsa_private_key

string

rsa_private_key

slack_token

string

slack_token

slack_webhook

string

slack_webhook

square_access_token

string

square_access_token

square_oauth_secret

string

square_oauth_secret

ssh_dsa_private_key

string

ssh_dsa_private_key

ssh_ec_private_key

string

ssh_ec_private_key

stripe_api_key

string

stripe_api_key

stripe_restricted_api_key

string

stripe_restricted_api_key

twilio_api_key

string

twilio_api_key

youtube_api_key

string

youtube_api_key

selfharm

object

disabled

boolean

disabled

action

string

action

threshold

number

minimum: 0
maximum: 1
multipleOf: 0.01

threshold

minimum: 0
maximum: 1
multipleOf: 0.01

sentiment

object

disabled

boolean

disabled

action

string

action

threshold

number

minimum: 0
maximum: 1
multipleOf: 0.01

threshold

minimum: 0
maximum: 1
multipleOf: 0.01

topic

object

disabled

boolean

disabled

action

string

(default: "")

action

topics

array<string>

Click 'Save' to use the value

threshold

number

minimum: 0
maximum: 1
multipleOf: 0.01

threshold

minimum: 0
maximum: 1
multipleOf: 0.01

log_fields

object

Additional fields to include in activity log.

citations

string

Origin or source application of the event. Examples: Slack, GDrive, Confluence, Workday.

citations

Origin or source application of the event. Examples: Slack, GDrive, Confluence, Workday.

extra_info

string

Stores supplementary details related to the event.

extra_info

Stores supplementary details related to the event.

model

string

Model used to perform the event. Example: {model: gpt, version: 3.5}.

model

Model used to perform the event. Example: {model: gpt, version: 3.5}.

source

string

IP address of user or app or agent.

source

IP address of user or app or agent.

tools

string

Tools used to perform the event. Example: {tool: ai-guard}.

tools

Tools used to perform the event. Example: {tool: ai-guard}.

Response Fields

Response Schema

object

Pangea standard response schema

result

object

prompt_text

string

Updated prompt text, if applicable.

prompt_messages

object

Updated structured prompt, if applicable.

blocked

boolean

Whether or not the prompt triggered a block detection.

recipe

string

The Recipe that was used.

detectors

object

Result of the recipe analyzing and input prompt.

prompt_injection

object

detected

boolean

Whether or not the Malicious Prompt was detected.

data

object

Details about the analyzers.

action

string

The action taken by this Detector

analyzer_responses

array<object>

Triggered prompt injection analyzers.

analyzer

string

confidence

number

gibberish

object

detected

boolean

Whether or not the Gibberish was detected.

data

object

Details about the analyzers.

action

string

The action taken by this Detector

classifications

array<object>

Triggered classifications.

category

string

confidence

number

sentiment

object

detected

boolean

Whether or not the Negative Sentiment was detected.

data

object

Details about the analyzers.

action

string

The action taken by this Detector

classifications

array<object>

Triggered classifications.

category

string

confidence

number

selfharm

object

detected

boolean

Whether or not the Self-harm was detected.

data

object

Details about the analyzers.

action

string

The action taken by this Detector

classifications

array<object>

Triggered classifications.

category

string

confidence

number

pii_entity

object

detected

boolean

Whether or not the PII Entities were detected.

data

object

Details about the detected entities.

entities

array<object>

Detected redaction rules.

action

string

The action taken on this Entity

type

string

value

string

redacted

boolean

start_pos

integer

minimum: 0

minimum: 0

malicious_entity

object

detected

boolean

Whether or not the Malicious Entities were detected.

data

object

Details about the detected entities.

entities

array<object>

Detected harmful items.

type

string

value

string

redacted

boolean

start_pos

integer

minimum: 0

minimum: 0

raw

object

custom_entity

object

detected

boolean

Whether or not the Custom Entities were detected.

data

object

Details about the detected entities.

entities

array<object>

Detected redaction rules.

action

string

The action taken on this Entity

type

string

value

string

redacted

boolean

start_pos

integer

minimum: 0

minimum: 0

secrets_detection

object

detected

boolean

Whether or not the Secret Entities were detected.

data

object

Details about the detected entities.

entities

array<object>

Detected redaction rules.

action

string

The action taken on this Entity

type

string

value

string

redacted

boolean

start_pos

integer

minimum: 0

minimum: 0

competitors

object

detected

boolean

Whether or not the Competitors were detected.

data

object

Details about the detected entities.

action

string

The action taken by this Detector

entities

array<string>

Detected entities.

profanity_and_toxicity

object

detected

boolean

Whether or not the Profanity and Toxic were detected.

data

object

Details about the detected entities.

action

string

The action taken by this Detector

classifications

array<object>

Triggered classifications.

category

string

confidence

number

hardening

object

data

object

Details about the detected languages.

action

string

The action taken by this Detector

message

string

Descriptive information about the hardening detector execution

token_count

number

Number of tokens counted in the last user prompt

language_detection

object

detected

boolean

Whether or not the Languages were detected.

data

object

Details about the detected languages.

action

string

The action taken by this Detector

language

string

topic

object

detected

boolean

Whether or not the Topics were detected.

data

object

Details about the detected topics.

action

string

The action taken by this Detector

topics

array<object>

List of topics detected

topic

string

confidence

number

code_detection

object

detected

boolean

Whether or not the Code was detected.

data

object

Details about the detected code.

action

string

The action taken by this Detector

language

string

fpe_context

string (base64)

If an FPE redaction method returned results, this will be the context passed to unredact.

request_id

string

A unique identifier assigned to each request made to the API. It is used to track and identify a specific request and its associated data. The request_id can be helpful for troubleshooting, auditing, and tracing the flow of requests within the system. It allows users to reference and retrieve information related to a particular request, such as the response, parameters, and raw data associated with that specific request.

"request_id":"prq_x6fdiizbon6j3bsdvnpmwxsz2aan7fqd"

"request_id":"prq_x6fdiizbon6j3bsdvnpmwxsz2aan7fqd"

request_time

string

The timestamp indicates the exact moment when a request is made to the API. It represents the date and time at which the request was initiated by the client. The request_time is useful for tracking and analyzing the timing of requests, measuring response times, and monitoring performance metrics. It allows users to determine the duration between the request initiation and the corresponding response, aiding in the assessment of API performance and latency.

"request_time":"2022-09-21T17:24:33.105Z"

"request_time":"2022-09-21T17:24:33.105Z"

response_time

string

Duration it takes for the API to process a request and generate a response. It represents the elapsed time from when the request is received by the API to when the corresponding response is returned to the client.

"response_time":"2022-09-21T17:24:34.007Z"

"response_time":"2022-09-21T17:24:34.007Z"

status

string

It represents the status or outcome of the API request made for IP information. It indicates the current state or condition of the request and provides information on the success or failure of the request.

"status":"success"

"status":"success"

summary

string

Provides a concise and brief overview of the purpose or primary objective of the API endpoint. It serves as a high-level summary or description of the functionality or feature offered by the endpoint.

post/v1beta/guard

curl -sSLX POST 'https://ai-guard.aws.us.pangea.cloud/v1beta/guard' \
-H 'Authorization: Bearer <your_token>' \
-H 'Content-Type: application/json' \
-d '{}'

Response

Raw Output

Guard LLM input and output

POST

https://ai-guard.aws.us.pangea.cloud/v1beta/guard

Analyze and redact content to avoid manipulation of the model, addition of malicious content, and other undesirable data transfers.

required parameters

messages

array<object>

Prompt content and role array in JSON format. The content is the multimodel text or image input that will be analyzed.

Click 'Save' to use the value

Prompt content and role array in JSON format. The content is the multimodel text or image input that will be analyzed.

optional parameters

recipe

string

(default: "pangea_prompt_guard")

Recipe key of a configuration of data types and settings defined in the Pangea User Console. It specifies the rules that are to be applied to the text, such as defang malicious URLs.

recipe

Recipe key of a configuration of data types and settings defined in the Pangea User Console. It specifies the rules that are to be applied to the text, such as defang malicious URLs.

debug

boolean

(default: false)

Setting this value to true will provide a detailed analysis of the text data

debug

Setting this value to true will provide a detailed analysis of the text data

overrides

object

ignore_recipe

boolean

(default: false)

Bypass existing Recipe content and create an on-the-fly Recipe.

ignore_recipe

Bypass existing Recipe content and create an on-the-fly Recipe.

code_detection

object

disabled

boolean

disabled

action

string

action

threshold

number

minimum: 0
maximum: 1
multipleOf: 0.01

threshold

minimum: 0
maximum: 1
multipleOf: 0.01

competitors

object

disabled

boolean

disabled

action

string

action

gibberish

object

disabled

boolean

disabled

action

string

action

language_detection

object

disabled

boolean

disabled

action

string

(default: "")

action

languages

array<string>

Click 'Save' to use the value

threshold

number

minimum: 0
maximum: 1
multipleOf: 0.01

threshold

minimum: 0
maximum: 1
multipleOf: 0.01

malicious_entity

object

disabled

boolean

disabled

domain

string

domain

ip_address

string

ip_address

url

string

url

pii_entity

object

disabled

boolean

disabled

au_abn

string

au_abn

au_acn

string

au_acn

au_medicare

string

au_medicare

au_tfn

string

au_tfn

credit_card

string

credit_card

crypto

string

crypto

date_time

string

date_time

email_address

string

email_address

fin/nric

string

fin/nric

iban_code

string

iban_code

ip_address

string

ip_address

location

string

location

medical_license

string

medical_license

money

string

money

nif

string

nif

nrp

string

nrp

person

string

person

phone_number

string

phone_number

uk_nhs

string

uk_nhs

url

string

url

us_bank_number

string

us_bank_number

us_drivers_license

string

us_drivers_license

us_itin

string

us_itin

us_passport

string

us_passport

us_ssn

string

us_ssn

prompt_injection

object

disabled

boolean

disabled

action

string

action

roleplay

object

disabled

boolean

disabled

action

string

action

secrets_detection

object

disabled

boolean

disabled

amazon_aws_access_key_id

string

amazon_aws_access_key_id

amazon_aws_secret_access_key

string

amazon_aws_secret_access_key

amazon_mws_auth_token

string

amazon_mws_auth_token

basic_auth

string

basic_auth

facebook_access_token

string

facebook_access_token

github_access_token

string

github_access_token

google_api_key

string

google_api_key

google_cloud_platform_api_key

string

google_cloud_platform_api_key

google_cloud_platform_service_account

string

google_cloud_platform_service_account

google_drive_api_key

string

google_drive_api_key

google_gmail_api_key

string

google_gmail_api_key

jwt_token

string

jwt_token

mailchimp_api_key

string

mailchimp_api_key

mailgun_api_key

string

mailgun_api_key

pangea_token

string

pangea_token

pgp_private_key_block

string

pgp_private_key_block

picatic_api_key

string

picatic_api_key

rsa_private_key

string

rsa_private_key

slack_token

string

slack_token

slack_webhook

string

slack_webhook

square_access_token

string

square_access_token

square_oauth_secret

string

square_oauth_secret

ssh_dsa_private_key

string

ssh_dsa_private_key

ssh_ec_private_key

string

ssh_ec_private_key

stripe_api_key

string

stripe_api_key

stripe_restricted_api_key

string

stripe_restricted_api_key

twilio_api_key

string

twilio_api_key

youtube_api_key

string

youtube_api_key

selfharm

object

disabled

boolean

disabled

action

string

action

threshold

number

minimum: 0
maximum: 1
multipleOf: 0.01

threshold

minimum: 0
maximum: 1
multipleOf: 0.01

sentiment

object

disabled

boolean

disabled

action

string

action

threshold

number

minimum: 0
maximum: 1
multipleOf: 0.01

threshold

minimum: 0
maximum: 1
multipleOf: 0.01

topic

object

disabled

boolean

disabled

action

string

(default: "")

action

topics

array<string>

Click 'Save' to use the value

threshold

number

minimum: 0
maximum: 1
multipleOf: 0.01

threshold

minimum: 0
maximum: 1
multipleOf: 0.01

app_name

string

Name of source application.

app_name

Name of source application.

llm_provider

string

Underlying LLM. Example: 'OpenAI'.

llm_provider

Underlying LLM. Example: 'OpenAI'.

model

string

Model used to perform the event. Example: 'gpt'.

model

Model used to perform the event. Example: 'gpt'.

model_version

string

Model version used to perform the event. Example: '3.5'.

model_version

Model version used to perform the event. Example: '3.5'.

request_token_count

integer

Number of tokens in the request.

request_token_count

Number of tokens in the request.

response_token_count

integer

Number of tokens in the response.

response_token_count

Number of tokens in the response.

source_ip

string

IP address of user or app or agent.

source_ip

IP address of user or app or agent.

source_location

string

Location of user or app or agent.

source_location

Location of user or app or agent.

tenant_id

string

For gateway-like integrations with multi-tenant support.

tenant_id

For gateway-like integrations with multi-tenant support.

sensor_mode

string

(AIDR) sensor mode.

sensor_mode

(AIDR) sensor mode.

context

object

(AIDR) Logging schema.

Click 'Save' to use the value

(AIDR) Logging schema.

Response Fields

Response Schema

object

Pangea standard response schema

result

object

prompt_messages

object

Updated structured prompt.

blocked

boolean

Whether or not the prompt triggered a block detection.

recipe

string

The Recipe that was used.

detectors

object

Result of the recipe analyzing and input prompt.

prompt_injection

object

detected

boolean

Whether or not the Malicious Prompt was detected.

data

object

Details about the analyzers.

action

string

The action taken by this Detector

analyzer_responses

array<object>

Triggered prompt injection analyzers.

analyzer

string

confidence

number

gibberish

object

detected

boolean

Whether or not the Gibberish was detected.

data

object

Details about the analyzers.

action

string

The action taken by this Detector

classifications

array<object>

Triggered classifications.

category

string

confidence

number

sentiment

object

detected

boolean

Whether or not the Negative Sentiment was detected.

data

object

Details about the analyzers.

action

string

The action taken by this Detector

classifications

array<object>

Triggered classifications.

category

string

confidence

number

selfharm

object

detected

boolean

Whether or not the Self-harm was detected.

data

object

Details about the analyzers.

action

string

The action taken by this Detector

classifications

array<object>

Triggered classifications.

category

string

confidence

number

pii_entity

object

detected

boolean

Whether or not the PII Entities were detected.

data

object

Details about the detected entities.

entities

array<object>

Detected redaction rules.

action

string

The action taken on this Entity

type

string

value

string

redacted

boolean

start_pos

integer

minimum: 0

minimum: 0

malicious_entity

object

detected

boolean

Whether or not the Malicious Entities were detected.

data

object

Details about the detected entities.

entities

array<object>

Detected harmful items.

type

string

value

string

redacted

boolean

start_pos

integer

minimum: 0

minimum: 0

raw

object

custom_entity

object

detected

boolean

Whether or not the Custom Entities were detected.

data

object

Details about the detected entities.

entities

array<object>

Detected redaction rules.

action

string

The action taken on this Entity

type

string

value

string

redacted

boolean

start_pos

integer

minimum: 0

minimum: 0

secrets_detection

object

detected

boolean

Whether or not the Secret Entities were detected.

data

object

Details about the detected entities.

entities

array<object>

Detected redaction rules.

action

string

The action taken on this Entity

type

string

value

string

redacted

boolean

start_pos

integer

minimum: 0

minimum: 0

competitors

object

detected

boolean

Whether or not the Competitors were detected.

data

object

Details about the detected entities.

action

string

The action taken by this Detector

entities

array<string>

Detected entities.

profanity_and_toxicity

object

detected

boolean

Whether or not the Profanity and Toxic were detected.

data

object

Details about the detected entities.

action

string

The action taken by this Detector

classifications

array<object>

Triggered classifications.

category

string

confidence

number

hardening

object

data

object

Details about the detected languages.

action

string

The action taken by this Detector

message

string

Descriptive information about the hardening detector execution

token_count

number

Number of tokens counted in the last user prompt

language_detection

object

detected

boolean

Whether or not the Languages were detected.

data

object

Details about the detected languages.

action

string

The action taken by this Detector

language

string

topic

object

detected

boolean

Whether or not the Topics were detected.

data

object

Details about the detected topics.

action

string

The action taken by this Detector

topics

array<object>

List of topics detected

topic

string

confidence

number

code_detection

object

detected

boolean

Whether or not the Code was detected.

data

object

Details about the detected code.

action

string

The action taken by this Detector

language

string

fpe_context

string (base64)

If an FPE redaction method returned results, this will be the context passed to unredact.

request_id

string

"request_id":"prq_x6fdiizbon6j3bsdvnpmwxsz2aan7fqd"

"request_id":"prq_x6fdiizbon6j3bsdvnpmwxsz2aan7fqd"

request_time

string

"request_time":"2022-09-21T17:24:33.105Z"

"request_time":"2022-09-21T17:24:33.105Z"

response_time

string

"response_time":"2022-09-21T17:24:34.007Z"

"response_time":"2022-09-21T17:24:34.007Z"

status

string

"status":"success"

"status":"success"

summary

string

Status Codes

Status	Status Code	Description
AIGBadInputFormat	400	The input schema is not currently supported.
AIGFileTooLarge	400	File size too large.
AIGInvalidRecipe	400	Recipe contains invalid contents.
AIGMissingSvConnection	400	A service dependency was not set up.
NoProviderConfigured	400	Provider for intelligence services used by ai-guard is not configured
RecipeNotExist	400	Recipe does not exist in the ai-guard configuration
AIGServiceConfigFetchError	500	Failed to fetch the dependent '{service_name}' service configuration from Pangea Cloud.

Was this article helpful?