fix(test): mock Lakera API in PII masking test for deterministic behavior

Changed from integration test to unit test by mocking the Lakera API response. This makes the test:
- Deterministic and not dependent on external API behavior
- Consistent with other tests in the file which are all mocked
- Able to test the masking logic regardless of Lakera's detection of test data
- Faster and more reliable in CI/CD

The mock response includes both credit card and email in the payload with proper start/end positions so we can verify the masking logic works correctly.

tests/guardrails_tests/test_lakera_v2.py

@@ -22,41 +22,58 @@ async def test_lakera_pre_call_hook_for_pii_masking():
     # Setup the guardrail with specific entities config
     litellm._turn_on_debug()
     lakera_guardrail = LakeraAIGuardrail(
-        api_key=os.environ.get("LAKERA_API_KEY"),
+        api_key="test_key",
     )
     
-    # Create a sample request with PII data
-    # Note: Using test email only, as test credit card numbers (like 4111-1111-1111-1111) 
-    # may not be consistently flagged by Lakera's API for masking in payload
-    data = {
-        "messages": [
-            {"role": "system", "content": "You are a helpful assistant."},
-            {"role": "user", "content": "My email is test@example.com and my phone number is 555-123-4567"}
+    # Mock response with PII detections in payload (with start/end positions for masking)
+    mock_response = {
+        'payload': [
+            {'detector_type': 'pii/credit_card', 'start': 18, 'end': 37, 'message_id': 1},  # "4111-1111-1111-1111"
+            {'detector_type': 'pii/email', 'start': 54, 'end': 70, 'message_id': 1},  # "test@example.com"
         ],
-        "model": "gpt-3.5-turbo",
-        "metadata": {}
+        'flagged': True,
+        'breakdown': [
+            {'detector_type': 'pii/credit_card', 'detected': True, 'message_id': 1},
+            {'detector_type': 'pii/email', 'detected': True, 'message_id': 1},
+        ]
     }
     
-    # Mock objects needed for the pre-call hook
-    user_api_key_dict = UserAPIKeyAuth(api_key="test_key")
-    cache = DualCache()
-    
-    # Call the pre-call hook with the specified call type
-    modified_data = await lakera_guardrail.async_pre_call_hook(
-        user_api_key_dict=user_api_key_dict,
-        cache=cache,
-        data=data,
-        call_type="completion"
-    )
-    print(modified_data)
-    
-    # Verify the messages have been modified to mask PII
-    assert modified_data["messages"][0]["content"] == "You are a helpful assistant."  # System prompt should be unchanged
-    
-    user_message = modified_data["messages"][1]["content"]
-    # Verify email is masked (Lakera should return this in payload for masking)
-    assert "test@example.com" not in user_message
-    assert "[MASKED EMAIL]" in user_message or "****" in user_message  # Accept either masking format
+    with patch.object(lakera_guardrail, 'call_v2_guard', new_callable=AsyncMock) as mock_call:
+        mock_call.return_value = (mock_response, {})
+        
+        # Create a sample request with PII data
+        data = {
+            "messages": [
+                {"role": "system", "content": "You are a helpful assistant."},
+                {"role": "user", "content": "My credit card is 4111-1111-1111-1111 and my email is test@example.com. My phone number is 555-123-4567"}
+            ],
+            "model": "gpt-3.5-turbo",
+            "metadata": {}
+        }
+        
+        # Mock objects needed for the pre-call hook
+        user_api_key_dict = UserAPIKeyAuth(api_key="test_key")
+        cache = DualCache()
+        
+        # Call the pre-call hook with the specified call type
+        modified_data = await lakera_guardrail.async_pre_call_hook(
+            user_api_key_dict=user_api_key_dict,
+            cache=cache,
+            data=data,
+            call_type="completion"
+        )
+        print(modified_data)
+        
+        # Verify the messages have been modified to mask PII
+        assert modified_data["messages"][0]["content"] == "You are a helpful assistant."  # System prompt should be unchanged
+        
+        user_message = modified_data["messages"][1]["content"]
+        # Verify both credit card and email are masked
+        assert "4111-1111-1111-1111" not in user_message
+        assert "test@example.com" not in user_message
+        # Verify masking placeholders are present
+        assert "[MASKED CREDIT_CARD]" in user_message
+        assert "[MASKED EMAIL]" in user_message
 
 
 @pytest.mark.asyncio