Additional foundry policies

policyAssignments/dev/pa-d-cog-service.json

@@ -70,6 +70,31 @@
         "value": [
           "FW-Kimi-K3"
         ]
+      },
+      "COG-010_Effect": {
+        "value": "Deny"
+      },
+      "COG-010_approvedRaiPolicies": {
+        "value": [
+          "Microsoft.DefaultV2",
+          "Contoso.DefaultV1"
+        ]
+      },
+      "COG-011_Effect": {
+        "value": "Deny"
+      },
+      "COG-011_modelNames": {
+        "value": [
+          "gpt-5-mini",
+          "gpt-5-nano",
+          "gpt-5.2",
+          "gpt-5.4-mini",
+          "gpt-5.4-nano",
+          "gpt-5.4",
+          "gpt-5.5",
+          "FW-Kimi-K3",
+          "FW-Kimi-K2.7-Code"
+        ]
       }
     },
     "nonComplianceMessages": [
@@ -80,6 +105,10 @@
       {
         "policyDefinitionReferenceId": "COG-007",
         "message": "PolicyID: COG-007 Violation in polset-cognitive-service Initiative - 'Only approved xAI models are allowed to be deployed in Cognitive Services'"
+      },
+      {
+        "policyDefinitionReferenceId": "COG-011",
+        "message": "PolicyID: COG-011 Violation in polset-cognitive-service Initiative - 'List of Foundry model deployments must use DataZoneStandard deployment type in AI Foundry accounts.'"
       }
     ],
     "roleDefinitionIds": [

policyAssignments/prod/pa-p-cog-service.json

@@ -70,6 +70,31 @@
         "value": [
           "FW-Kimi-K3"
         ]
+      },
+      "COG-010_Effect": {
+        "value": "Deny"
+      },
+      "COG-010_approvedRaiPolicies": {
+        "value": [
+          "Microsoft.DefaultV2",
+          "Contoso.DefaultV1"
+        ]
+      },
+      "COG-011_Effect": {
+        "value": "Deny"
+      },
+      "COG-011_modelNames": {
+        "value": [
+          "gpt-5-mini",
+          "gpt-5-nano",
+          "gpt-5.2",
+          "gpt-5.4-mini",
+          "gpt-5.4-nano",
+          "gpt-5.4",
+          "gpt-5.5",
+          "FW-Kimi-K3",
+          "FW-Kimi-K2.7-Code"
+        ]
       }
     },
     "nonComplianceMessages": [
@@ -80,6 +105,10 @@
       {
         "policyDefinitionReferenceId": "COG-007",
         "message": "PolicyID: COG-007 Violation in polset-cognitive-service Initiative - 'Only approved xAI models are allowed to be deployed in Cognitive Services'"
+      },
+      {
+        "policyDefinitionReferenceId": "COG-011",
+        "message": "PolicyID: COG-011 Violation in polset-cognitive-service Initiative - 'List of Foundry model deployments must use DataZoneStandard deployment type in AI Foundry accounts.'"
       }
     ],
     "roleDefinitionIds": [

policyDefinitions/cognitive-service/pol-enforce-foundry-deployment-rai-policy.json

@@ -0,0 +1,66 @@
+{
+  "name": "pol-enforce-foundry-deployment-rai-policy",
+  "properties": {
+    "displayName": "Foundry deployments must have an approved Responsible AI policy assigned",
+    "description": "Ensure Foundry model deployments have an associated approved Responsible AI policy.",
+    "mode": "All",
+    "metadata": {
+      "version": "1.0.0",
+      "category": "Cognitive Services",
+      "preview": false,
+      "deprecated": false
+    },
+    "parameters": {
+      "effect": {
+        "type": "String",
+        "metadata": {
+          "displayName": "Effect",
+          "description": "Enable or disable execution of the policy."
+        },
+        "allowedValues": [
+          "Audit",
+          "Deny",
+          "Disabled"
+        ],
+        "defaultValue": "Audit"
+      },
+      "approvedRaiPolicies": {
+        "type": "Array",
+        "metadata": {
+          "displayName": "Approved RAI policy names",
+          "description": "Responsible AI policy names that satisfy the organisation content safety baseline."
+        },
+        "defaultValue": [
+          "Microsoft.DefaultV2"
+        ]
+      }
+    },
+    "policyRule": {
+      "if": {
+        "allOf": [
+          {
+            "field": "type",
+            "equals": "Microsoft.CognitiveServices/accounts/deployments"
+          },
+          {
+            "anyOf": [
+              {
+                "field": "Microsoft.CognitiveServices/accounts/deployments/raiPolicyName",
+                "exists": "false"
+              },
+              {
+                "not": {
+                  "field": "Microsoft.CognitiveServices/accounts/deployments/raiPolicyName",
+                  "in": "[parameters('approvedRaiPolicies')]"
+                }
+              }
+            ]
+          }
+        ]
+      },
+      "then": {
+        "effect": "[parameters('effect')]"
+      }
+    }
+  }
+}

policyDefinitions/cognitive-service/pol-enforce-foundry-model-deployment-type.json

@@ -0,0 +1,63 @@
+{
+  "name": "pol-enforce-foundry-model-deployment-type",
+  "properties": {
+    "displayName": "List of Foundry Model deployments must use approved deployment type",
+    "mode": "All",
+    "description": "Enforce Foundry models to be deployed using approved deployment types.",
+    "metadata": {
+      "version": "1.0.0",
+      "category": "Cognitive Services"
+    },
+    "parameters": {
+      "effect": {
+        "type": "String",
+        "metadata": {
+          "displayName": "Effect",
+          "description": "Enable or disable execution of the policy."
+        },
+        "allowedValues": [
+          "Audit",
+          "Deny",
+          "Disabled"
+        ],
+        "defaultValue": "Deny"
+      },
+      "modelNames": {
+        "type": "Array",
+        "metadata": {
+          "displayName": "In-scope model names",
+          "description": "Model names this restriction applies to. Matched against model.name. Comparison is case-insensitive."
+        }
+      },
+      "allowedDeploymentType": {
+        "type": "string",
+        "metadata": {
+          "displayName": "Approved deployment type",
+          "description": "Permitted value of sku.name for the in-scope models."
+        },
+        "defaultValue": "DataZoneStandard"
+      }
+    },
+    "policyRule": {
+      "if": {
+        "allOf": [
+          {
+            "field": "type",
+            "equals": "Microsoft.CognitiveServices/accounts/deployments"
+          },
+          {
+            "field": "Microsoft.CognitiveServices/accounts/deployments/model.name",
+            "in": "[parameters('modelNames')]"
+          },
+          {
+            "field": "Microsoft.CognitiveServices/accounts/deployments/sku.name",
+            "notEquals": "[parameters('allowedDeploymentType')]"
+          }
+        ]
+      },
+      "then": {
+        "effect": "[parameters('effect')]"
+      }
+    }
+  }
+}

policyInitiatives/polset-cognitive-service.json

@@ -165,6 +165,53 @@
           "displayName": "COG-009 Allowed Model Names for Fireworks format",
           "description": "List of Model Names that are allowed to be deployed."
         }
+      },
+      "COG-010_Effect": {
+        "type": "String",
+        "metadata": {
+          "displayName": "COG-010 Effect: Foundry deployments must have an approved Responsible AI policy assigned",
+          "description": "'Audit' allows a non-compliant resource to be created, but flags it as non-compliant. 'Deny' blocks the resource creation. 'Disable' turns off the policy."
+        },
+        "allowedValues": [
+          "Audit",
+          "Deny",
+          "Disabled"
+        ],
+        "defaultValue": "Deny"
+      },
+      "COG-010_approvedRaiPolicies": {
+        "type": "Array",
+        "metadata": {
+          "displayName": "COG-010 Approved RAI policy names",
+          "description": "Responsible AI policy names that satisfy the organisation content safety baseline."
+        },
+        "defaultValue": [
+          "Microsoft.DefaultV2"
+        ]
+      },
+      "COG-011_Effect": {
+        "type": "String",
+        "metadata": {
+          "displayName": "COG-011 Effect: List of Foundry Model deployments must use DataZoneStandard deployment type",
+          "description": "'Audit' allows a non-compliant resource to be created, but flags it as non-compliant. 'Deny' blocks the resource creation. 'Disable' turns off the policy."
+        },
+        "allowedValues": [
+          "Audit",
+          "Deny",
+          "Disabled"
+        ],
+        "defaultValue": "Deny"
+      },
+      "COG-011_modelNames": {
+        "type": "Array",
+        "metadata": {
+          "displayName": "COG-011 In-scope model names",
+          "description": "Model names this restriction applies to. Matched against model.name. Comparison is case-insensitive."
+        },
+        "defaultValue": [
+          "gpt-5-mini",
+          "gpt-5-nano"
+        ]
       }
     },
     "policyDefinitionGroups": [
@@ -195,6 +242,14 @@
       {
         "name": "ISO27001-2013_A.14.2.4",
         "additionalMetadataId": "/providers/Microsoft.PolicyInsights/policyMetadata/ISO27001-2013_A.14.2.4"
+      },
+      {
+        "name": "ISO27001-2013_A.14.2.5",
+        "additionalMetadataId": "/providers/Microsoft.PolicyInsights/policyMetadata/ISO27001-2013_A.14.2.5"
+      },
+      {
+        "name": "ISO27001-2013_A.18.1.1",
+        "additionalMetadataId": "/providers/Microsoft.PolicyInsights/policyMetadata/ISO27001-2013_A.18.1.1"
       }
     ],
     "policyDefinitions": [
@@ -341,6 +396,39 @@
           "ISO27001-2013_A.12.5.1",
           "ISO27001-2013_A.14.2.4"
         ]
+      },
+      {
+        "policyDefinitionReferenceId": "COG-010",
+        "policyDefinitionId": "{policyLocationResourceId}/providers/Microsoft.Authorization/policyDefinitions/pol-enforce-foundry-deployment-rai-policy",
+        "parameters": {
+          "effect": {
+            "value": "[parameters('COG-010_Effect')]"
+          },
+          "approvedRaiPolicies": {
+            "value": "[parameters('COG-010_approvedRaiPolicies')]"
+          }
+        },
+        "groupNames": [
+          "ISO27001-2013_A.14.2.5"
+        ]
+      },
+      {
+        "policyDefinitionReferenceId": "COG-011",
+        "policyDefinitionId": "{policyLocationResourceId}/providers/Microsoft.Authorization/policyDefinitions/pol-enforce-foundry-model-deployment-type",
+        "parameters": {
+          "effect": {
+            "value": "[parameters('COG-011_Effect')]"
+          },
+          "modelNames": {
+            "value": "[parameters('COG-011_modelNames')]"
+          },
+          "allowedDeploymentType": {
+            "value": "DataZoneStandard"
+          }
+        },
+        "groupNames": [
+          "ISO27001-2013_A.18.1.1"
+        ]
       }
     ]
   }

tests/policy-integration-tests/cognitive-service/README.md

@@ -23,5 +23,7 @@ The following policies are in scope for testing:
 | `pa-d-cog-service` | `COG-005` | Cognitive Services Deployments allowed model formats | Deny |
 | `pa-d-cog-service` | `COG-006` | Cognitive Services Deployments should only use approved Models from OpenAI | Deny |
 | `pa-d-cog-service` | `COG-007` | Cognitive Services Deployments should only use approved Models from xAI | Deny |
+| `pa-d-cog-service` | `COG-010` | Cognitive Services Deployments should use an allowed Responsible AI (RAI) policy | Deny |
+| `pa-d-cog-service` | `COG-011` | List of Foundry Model deployments must use DataZoneStandard deployment type | Deny |
 | `pa-d-diag-settings` | `DS-013` | Configure Diagnostic Setting for Azure Cognitive Services | DeployIfNotExists |
 | `pa-d-pedns` | `PEDNS-016` | Private DNS Record for Azure Cognitive Services PE must exist | DeployIfNotExists |

tests/policy-integration-tests/cognitive-service/main.bad.bicep

@@ -45,6 +45,21 @@ resource gpt51 'Microsoft.CognitiveServices/accounts/deployments@2025-12-01' = {
       name: 'gpt-5.1' //this should violate the policy COG-006 since gpt-5.1 is not in the allowed list of models defined in the policy
       format: 'OpenAI'
     }
+    raiPolicyName: 'rai-policy-1' //this should violate the policy COG-010 since rai-policy-1 is not in the allowed list of RAI policies defined in the policy
+  }
+}
+resource gpt54 'Microsoft.CognitiveServices/accounts/deployments@2025-12-01' = {
+  name: 'gpt54'
+  parent: cognitiveService
+  sku: {
+    name: 'GlobalStandard' //this should violate the policy COG-011 since GlobalStandard is not allowed for this model
+    capacity: 150
+  }
+  properties: {
+    model: {
+      name: 'gpt-5.4'
+      format: 'OpenAI'
+    }
   }
 }
 

tests/policy-integration-tests/cognitive-service/main.good.bicep

@@ -39,18 +39,19 @@ resource cognitiveService 'Microsoft.CognitiveServices/accounts@2025-12-01' = {
     ] //user owned storage defined, this should comply with the policy COG-004
   }
 }
-resource gpt41 'Microsoft.CognitiveServices/accounts/deployments@2025-12-01' = {
-  name: 'gpt41'
+resource gpt54 'Microsoft.CognitiveServices/accounts/deployments@2025-12-01' = {
+  name: 'gpt54'
   parent: cognitiveService
   sku: {
-    name: 'GlobalStandard'
+    name: 'DataZoneStandard' //this should comply with the policy COG-011 since this model must use DataZoneStandard as the deployment type
     capacity: 150
   }
   properties: {
     model: {
       name: 'gpt-5.4'
       format: 'OpenAI'
     }
+    raiPolicyName: 'Microsoft.DefaultV2' //this should comply with the policy COG-010 since Microsoft.DefaultV2 is in the allowed list of RAI policies defined in the policy
   }
 }
 
@@ -67,6 +68,7 @@ resource grok 'Microsoft.CognitiveServices/accounts/deployments@2025-12-01' = {
       format: 'xAI'
       version: '1'
     }
+    raiPolicyName: 'Microsoft.DefaultV2' //this should comply with the policy COG-010 since Contoso.DefaultV1 is in the allowed list of RAI policies defined in the policy
   }
 }