diff --git a/nemo/collections/common/metrics/classification_accuracy.py b/nemo/collections/common/metrics/classification_accuracy.py
index aa14d5c320db..c7d7e1373434 100644
--- a/nemo/collections/common/metrics/classification_accuracy.py
+++ b/nemo/collections/common/metrics/classification_accuracy.py
@@ -161,8 +161,10 @@ def compute_topk_accuracy(correct_counts_k, total_counts_k):
 
 
 class ExactStringPerCategoryMatchMetric(Metric):
-    def __init__(self, categories=[], dist_sync_on_step=False, *args, **kwargs):
+    def __init__(self, categories=None, dist_sync_on_step=False, *args, **kwargs):
         super().__init__(dist_sync_on_step=dist_sync_on_step)
+        if categories is None:
+            categories = []
         self.categories = set(categories)
 
         self.add_state("correct", default=torch.tensor(0), dist_reduce_fx="sum")
diff --git a/tests/collections/common/test_metrics.py b/tests/collections/common/test_metrics.py
index 64ab9a4048af..977d302b3305 100644
--- a/tests/collections/common/test_metrics.py
+++ b/tests/collections/common/test_metrics.py
@@ -15,7 +15,10 @@
 import pytest
 import torch
 
-from nemo.collections.common.metrics.classification_accuracy import TopKClassificationAccuracy
+from nemo.collections.common.metrics.classification_accuracy import (
+    ExactStringPerCategoryMatchMetric,
+    TopKClassificationAccuracy,
+)
 from nemo.collections.common.metrics.punct_er import (
     DatasetPunctuationErrorRate,
     OccurancePunctuationErrorRate,
@@ -86,6 +89,26 @@ def test_top_1_accuracy_distributed(self):
 
         assert abs(acc_top1 - 0.5) < 1e-3  # 3/6
 
+    @pytest.mark.unit
+    def test_exact_string_per_category_match_metric_default_categories_is_none(self):
+        """Default categories should be None to avoid shared mutable state."""
+        import inspect
+
+        sig = inspect.signature(ExactStringPerCategoryMatchMetric.__init__)
+        assert sig.parameters['categories'].default is None
+
+    @pytest.mark.unit
+    def test_exact_string_per_category_match_metric_tracks_categories(self):
+        """Metric should track per-category correct/total counts."""
+        metric = ExactStringPerCategoryMatchMetric(categories=['a', 'b'])
+        metric.update(pred='x', target='x', category='a')
+        metric.update(pred='x', target='y', category='b')
+
+        assert metric.a_total == 1
+        assert metric.a_correct == 1
+        assert metric.b_total == 1
+        assert metric.b_correct == 0
+
     @pytest.mark.unit
     def test_top_1_accuracy_distributed_uneven_batch(self):
         # Simulate test on 2 process DDP execution