Document AI Guard evaluation API

examples/v2/ai-guard/EvaluateAIGuardRequest.rb

@@ -0,0 +1,26 @@
+# Evaluate an AI Guard request returns "OK" response
+
+require "datadog_api_client"
+DatadogAPIClient.configure do |config|
+  config.access_token = ENV["DD_BEARER_TOKEN"]
+  config.unstable_operations["v2.evaluate_ai_guard_request".to_sym] = true
+end
+api_instance = DatadogAPIClient::V2::AIGuardAPI.new
+
+body = DatadogAPIClient::V2::AIGuardEvaluateRequest.new({
+  data: DatadogAPIClient::V2::AIGuardEvaluateRequestData.new({
+    type: DatadogAPIClient::V2::AIGuardEvaluationType::EVALUATIONS,
+    attributes: DatadogAPIClient::V2::AIGuardEvaluateRequestDataAttributes.new({
+      messages: [
+        DatadogAPIClient::V2::AIGuardMessage.new({
+          content: "Hello world!",
+          role: DatadogAPIClient::V2::AIGuardMessageRole::USER,
+        }),
+      ],
+      meta: DatadogAPIClient::V2::AIGuardMeta.new({
+        confidence_threshold: 0.8,
+      }),
+    }),
+  }),
+})
+p api_instance.evaluate_ai_guard_request(body)

examples/v2/ai-guard/EvaluateAIGuardRequest_1377048937.rb

@@ -0,0 +1,29 @@
+# Evaluate an AI Guard request with per-category confidence thresholds
+
+require "datadog_api_client"
+DatadogAPIClient.configure do |config|
+  config.access_token = ENV["DD_BEARER_TOKEN"]
+  config.unstable_operations["v2.evaluate_ai_guard_request".to_sym] = true
+end
+api_instance = DatadogAPIClient::V2::AIGuardAPI.new
+
+body = DatadogAPIClient::V2::AIGuardEvaluateRequest.new({
+  data: DatadogAPIClient::V2::AIGuardEvaluateRequestData.new({
+    type: DatadogAPIClient::V2::AIGuardEvaluationType::EVALUATIONS,
+    attributes: DatadogAPIClient::V2::AIGuardEvaluateRequestDataAttributes.new({
+      messages: [
+        DatadogAPIClient::V2::AIGuardMessage.new({
+          role: DatadogAPIClient::V2::AIGuardMessageRole::USER,
+          content: "Hello world!",
+        }),
+      ],
+      meta: DatadogAPIClient::V2::AIGuardMeta.new({
+        confidence_threshold: DatadogAPIClient::V2::AIGuardConfidenceThreshold.new({
+          flagging_global: 0.6,
+          "policy-evasion" => 0.9, "indirect-prompt-injection" => 0.4,
+        }),
+      }),
+    }),
+  }),
+})
+p api_instance.evaluate_ai_guard_request(body)

examples/v2/ai-guard/EvaluateAIGuardRequest_3923646360.rb

@@ -0,0 +1,26 @@
+# Evaluate an AI Guard request with a numeric confidence threshold
+
+require "datadog_api_client"
+DatadogAPIClient.configure do |config|
+  config.access_token = ENV["DD_BEARER_TOKEN"]
+  config.unstable_operations["v2.evaluate_ai_guard_request".to_sym] = true
+end
+api_instance = DatadogAPIClient::V2::AIGuardAPI.new
+
+body = DatadogAPIClient::V2::AIGuardEvaluateRequest.new({
+  data: DatadogAPIClient::V2::AIGuardEvaluateRequestData.new({
+    type: DatadogAPIClient::V2::AIGuardEvaluationType::EVALUATIONS,
+    attributes: DatadogAPIClient::V2::AIGuardEvaluateRequestDataAttributes.new({
+      messages: [
+        DatadogAPIClient::V2::AIGuardMessage.new({
+          role: DatadogAPIClient::V2::AIGuardMessageRole::USER,
+          content: "Hello world!",
+        }),
+      ],
+      meta: DatadogAPIClient::V2::AIGuardMeta.new({
+        confidence_threshold: 0.8,
+      }),
+    }),
+  }),
+})
+p api_instance.evaluate_ai_guard_request(body)

features/scenarios_model_mapping.rb

@@ -1340,6 +1340,9 @@
     "v2.GetAwsOnDemandTask" => {
             "task_id" => "String",
     },
+    "v2.EvaluateAIGuardRequest" => {
+            "body" => "AIGuardEvaluateRequest",
+    },
     "v2.CreateAIImpactUserActivity" => {
             "body" => "AIImpactUserActivityRequest",
     },

features/v2/ai_guard.feature

@@ -0,0 +1,71 @@
+@endpoint(ai-guard) @endpoint(ai-guard-v2)
+Feature: AI Guard
+  Evaluate LLM interactions for security threats such as prompt injection
+  and jailbreak attempts.
+
+  Background:
+    Given a valid "apiKeyAuth" key in the system
+    And a valid "appKeyAuth" key in the system
+    And an instance of "AIGuard" API
+    And operation "EvaluateAIGuardRequest" enabled
+    And new "EvaluateAIGuardRequest" request
+
+  @team:ddoghq/k9-ai-guard
+  Scenario: Evaluate an AI Guard request rejects empty multipart user content
+    Given body with value {"data":{"type":"evaluations","attributes":{"messages":[{"role":"user","content":[]}]}}}
+    When the request is sent
+    Then the response status is 400 Bad Request
+
+  @team:ddoghq/k9-ai-guard
+  Scenario: Evaluate an AI Guard request rejects image parts without a URL
+    Given body with value {"data":{"type":"evaluations","attributes":{"messages":[{"role":"user","content":[{"type":"image_url","image_url":{"url":""}}]}]}}}
+    When the request is sent
+    Then the response status is 400 Bad Request
+
+  @team:ddoghq/k9-ai-guard
+  Scenario: Evaluate an AI Guard request rejects text parts without text
+    Given body with value {"data":{"type":"evaluations","attributes":{"messages":[{"role":"user","content":[{"type":"text"}]}]}}}
+    When the request is sent
+    Then the response status is 400 Bad Request
+
+  @team:ddoghq/k9-ai-guard
+  Scenario: Evaluate an AI Guard request returns "Bad Request" response
+    Given body with value {"data":{"type":"evaluations","attributes":{"messages":[{"role":"user"}]}}}
+    When the request is sent
+    Then the response status is 400 Bad Request
+
+  @team:ddoghq/k9-ai-guard
+  Scenario: Evaluate an AI Guard request returns "OK" response
+    Given body with value {"data": {"type": "evaluations", "attributes": {"messages": [{"content": "Hello world!", "role": "user"}], "meta": {"confidence_threshold": 0.8}}}}
+    When the request is sent
+    Then the response status is 200 OK
+
+  @generated @skip @team:ddoghq/k9-ai-guard
+  Scenario: Evaluate an AI Guard request returns "Payload Too Large" response
+    Given body with value {"data": {"attributes": {"messages": [{"content": "Hello world!", "role": "user"}], "meta": {"confidence_threshold": 0.8}}, "type": "evaluations"}}
+    When the request is sent
+    Then the response status is 413 Payload Too Large
+
+  @generated @skip @team:ddoghq/k9-ai-guard
+  Scenario: Evaluate an AI Guard request returns "Request Timeout" response
+    Given body with value {"data": {"attributes": {"messages": [{"content": "Hello world!", "role": "user"}], "meta": {"confidence_threshold": 0.8}}, "type": "evaluations"}}
+    When the request is sent
+    Then the response status is 408 Request Timeout
+
+  @team:ddoghq/k9-ai-guard
+  Scenario: Evaluate an AI Guard request with a numeric confidence threshold
+    Given body with value {"data":{"type":"evaluations","attributes":{"messages":[{"role":"user","content":"Hello world!"}],"meta":{"confidence_threshold":0.8}}}}
+    When the request is sent
+    Then the response status is 200 OK
+    And the response "data.type" is equal to "evaluations"
+    And the response "data.attributes" has field "action"
+    And the response "data.attributes" has field "is_blocking_enabled"
+
+  @team:ddoghq/k9-ai-guard
+  Scenario: Evaluate an AI Guard request with per-category confidence thresholds
+    Given body with value {"data":{"type":"evaluations","attributes":{"messages":[{"role":"user","content":"Hello world!"}],"meta":{"confidence_threshold":{"flagging_global":0.6,"policy-evasion":0.9,"indirect-prompt-injection":0.4}}}}}
+    When the request is sent
+    Then the response status is 200 OK
+    And the response "data.type" is equal to "evaluations"
+    And the response "data.attributes" has field "action"
+    And the response "data.attributes" has field "is_blocking_enabled"

features/v2/undo.json

@@ -374,6 +374,12 @@
       "type": "safe"
     }
   },
+  "EvaluateAIGuardRequest": {
+    "tag": "AI Guard",
+    "undo": {
+      "type": "safe"
+    }
+  },
   "CreateAIImpactUserActivity": {
     "tag": "Delivery Performance",
     "undo": {

generated-test/test-runner-data/manifest.json

@@ -2793,6 +2793,55 @@
       "scenario": "Update GCP scan options returns \"OK\" response",
       "version": "v2"
     },
+    {
+      "feature": "AI Guard",
+      "feature_file": "../../features/v2/ai_guard.feature",
+      "file": "v2/ai-guard/evaluate-an-ai-guard-request-rejects-empty-multipart-user-content.json",
+      "scenario": "Evaluate an AI Guard request rejects empty multipart user content",
+      "version": "v2"
+    },
+    {
+      "feature": "AI Guard",
+      "feature_file": "../../features/v2/ai_guard.feature",
+      "file": "v2/ai-guard/evaluate-an-ai-guard-request-rejects-image-parts-without-a-url.json",
+      "scenario": "Evaluate an AI Guard request rejects image parts without a URL",
+      "version": "v2"
+    },
+    {
+      "feature": "AI Guard",
+      "feature_file": "../../features/v2/ai_guard.feature",
+      "file": "v2/ai-guard/evaluate-an-ai-guard-request-rejects-text-parts-without-text.json",
+      "scenario": "Evaluate an AI Guard request rejects text parts without text",
+      "version": "v2"
+    },
+    {
+      "feature": "AI Guard",
+      "feature_file": "../../features/v2/ai_guard.feature",
+      "file": "v2/ai-guard/evaluate-an-ai-guard-request-returns-bad-request-response.json",
+      "scenario": "Evaluate an AI Guard request returns \"Bad Request\" response",
+      "version": "v2"
+    },
+    {
+      "feature": "AI Guard",
+      "feature_file": "../../features/v2/ai_guard.feature",
+      "file": "v2/ai-guard/evaluate-an-ai-guard-request-returns-ok-response.json",
+      "scenario": "Evaluate an AI Guard request returns \"OK\" response",
+      "version": "v2"
+    },
+    {
+      "feature": "AI Guard",
+      "feature_file": "../../features/v2/ai_guard.feature",
+      "file": "v2/ai-guard/evaluate-an-ai-guard-request-with-a-numeric-confidence-threshold.json",
+      "scenario": "Evaluate an AI Guard request with a numeric confidence threshold",
+      "version": "v2"
+    },
+    {
+      "feature": "AI Guard",
+      "feature_file": "../../features/v2/ai_guard.feature",
+      "file": "v2/ai-guard/evaluate-an-ai-guard-request-with-per-category-confidence-thresholds.json",
+      "scenario": "Evaluate an AI Guard request with per-category confidence thresholds",
+      "version": "v2"
+    },
     {
       "feature": "Annotations",
       "feature_file": "../../features/v2/annotations.feature",

generated-test/test-runner-data/v2/ai-guard/evaluate-an-ai-guard-request-rejects-empty-multipart-user-content.json

@@ -0,0 +1,38 @@
+{
+  "api": "AIGuard",
+  "expected_status": 400,
+  "feature": "AI Guard",
+  "id": "v2/AI Guard/Evaluate an AI Guard request rejects empty multipart user content",
+  "operation_id": "EvaluateAIGuardRequest",
+  "request": {
+    "body": {
+      "schema": {
+        "format": null,
+        "ref": "AIGuardEvaluateRequest",
+        "type": "object"
+      },
+      "source": "inline",
+      "value": {
+        "data": {
+          "attributes": {
+            "messages": [
+              {
+                "content": [],
+                "role": "user"
+              }
+            ]
+          },
+          "type": "evaluations"
+        }
+      }
+    },
+    "content_type": "application/json",
+    "method": "POST",
+    "pagination": false,
+    "parameters": [],
+    "path": "/api/v2/ai-guard/evaluate"
+  },
+  "scenario": "Evaluate an AI Guard request rejects empty multipart user content",
+  "schema_version": 1,
+  "version": "v2"
+}

generated-test/test-runner-data/v2/ai-guard/evaluate-an-ai-guard-request-rejects-image-parts-without-a-url.json

@@ -0,0 +1,45 @@
+{
+  "api": "AIGuard",
+  "expected_status": 400,
+  "feature": "AI Guard",
+  "id": "v2/AI Guard/Evaluate an AI Guard request rejects image parts without a URL",
+  "operation_id": "EvaluateAIGuardRequest",
+  "request": {
+    "body": {
+      "schema": {
+        "format": null,
+        "ref": "AIGuardEvaluateRequest",
+        "type": "object"
+      },
+      "source": "inline",
+      "value": {
+        "data": {
+          "attributes": {
+            "messages": [
+              {
+                "content": [
+                  {
+                    "image_url": {
+                      "url": ""
+                    },
+                    "type": "image_url"
+                  }
+                ],
+                "role": "user"
+              }
+            ]
+          },
+          "type": "evaluations"
+        }
+      }
+    },
+    "content_type": "application/json",
+    "method": "POST",
+    "pagination": false,
+    "parameters": [],
+    "path": "/api/v2/ai-guard/evaluate"
+  },
+  "scenario": "Evaluate an AI Guard request rejects image parts without a URL",
+  "schema_version": 1,
+  "version": "v2"
+}

generated-test/test-runner-data/v2/ai-guard/evaluate-an-ai-guard-request-rejects-text-parts-without-text.json

@@ -0,0 +1,42 @@
+{
+  "api": "AIGuard",
+  "expected_status": 400,
+  "feature": "AI Guard",
+  "id": "v2/AI Guard/Evaluate an AI Guard request rejects text parts without text",
+  "operation_id": "EvaluateAIGuardRequest",
+  "request": {
+    "body": {
+      "schema": {
+        "format": null,
+        "ref": "AIGuardEvaluateRequest",
+        "type": "object"
+      },
+      "source": "inline",
+      "value": {
+        "data": {
+          "attributes": {
+            "messages": [
+              {
+                "content": [
+                  {
+                    "type": "text"
+                  }
+                ],
+                "role": "user"
+              }
+            ]
+          },
+          "type": "evaluations"
+        }
+      }
+    },
+    "content_type": "application/json",
+    "method": "POST",
+    "pagination": false,
+    "parameters": [],
+    "path": "/api/v2/ai-guard/evaluate"
+  },
+  "scenario": "Evaluate an AI Guard request rejects text parts without text",
+  "schema_version": 1,
+  "version": "v2"
+}