diff --git a/app/ai-service/conftest.py b/app/ai-service/conftest.py
index e1428362..71a9640f 100644
--- a/app/ai-service/conftest.py
+++ b/app/ai-service/conftest.py
@@ -44,11 +44,8 @@ def _make_pkg(name: str):
         if _mod not in sys.modules:
             sys.modules[_mod] = _make_pkg(_mod)
 
-# proof_of_life raises RuntimeError at import time when cv2 is mocked.
-_pol = _make_pkg("proof_of_life")
-_pol.ProofOfLifeAnalyzer = MagicMock()
-_pol.ProofOfLifeConfig = MagicMock()
-sys.modules["proof_of_life"] = _pol
+# proof_of_life's __init__ loads cv2 cascade classifiers which work fine
+# with the mocked cv2 in conftest — no module-level stub needed.
 
 # Patch metrics.check_system_resources so the monitor_requests middleware
 # doesn't crash when torch (vram) is a MagicMock.
diff --git a/app/ai-service/proof_of_life.py b/app/ai-service/proof_of_life.py
index 854bf524..ea0837b1 100644
--- a/app/ai-service/proof_of_life.py
+++ b/app/ai-service/proof_of_life.py
@@ -120,13 +120,16 @@ def analyze(
 
         burst_required = bool(burst_images_base64)
         has_liveness_evidence = (
-            checks["blink_detected"] or checks["head_movement_detected"] or not burst_required
+            checks["blink_detected"] or checks["head_movement_detected"]
         )
         is_real_person = confidence >= threshold and has_liveness_evidence
 
         reason = "Face detected and confidence threshold met"
         if burst_required and not has_liveness_evidence:
             reason = "No liveness signal detected from burst frames"
+        elif not burst_required:
+            is_real_person = False
+            reason = "Liveness verification requires burst frames"
         elif confidence < threshold:
             reason = "Confidence score is below threshold"
 
diff --git a/app/ai-service/tests/test_proof_of_life.py b/app/ai-service/tests/test_proof_of_life.py
new file mode 100644
index 00000000..f27c7f44
--- /dev/null
+++ b/app/ai-service/tests/test_proof_of_life.py
@@ -0,0 +1,101 @@
+"""Tests for proof-of-life liveness gate (Issue #431).
+
+Verifies that:
+  - A selfie-only request (no burst frames) returns is_real_person: false.
+  - Burst-based requests are scored on actual blink/head-movement evidence.
+"""
+
+from unittest.mock import patch, MagicMock
+
+import numpy as np
+
+from proof_of_life import ProofOfLifeAnalyzer, ProofOfLifeConfig
+
+
+def _make_analyzer():
+    """Build an analyzer with mocked cascade classifiers."""
+    cfg = ProofOfLifeConfig(confidence_threshold=0.65)
+    with patch("cv2.CascadeClassifier") as mock_cls:
+        mock_instance = MagicMock()
+        mock_instance.empty.return_value = False
+        mock_cls.return_value = mock_instance
+        analyzer = ProofOfLifeAnalyzer(config=cfg)
+    return analyzer
+
+
+def _fake_decode(image_base64: str) -> np.ndarray:
+    """Return a synthetic 200x200 BGR image for any base64 input."""
+    return np.zeros((200, 200, 3), dtype=np.uint8)
+
+
+class TestSelfieOnlyRefusal:
+    """Selfie-only requests must always be refused."""
+
+    def test_selfie_only_returns_false(self):
+        analyzer = _make_analyzer()
+        with patch.object(analyzer, "_decode_image", side_effect=_fake_decode), \
+             patch.object(analyzer, "_detect_primary_face", return_value=(50, 50, 100, 100)):
+            result = analyzer.analyze(selfie_image_base64="dGVzdA==")
+        assert result["is_real_person"] is False
+
+    def test_selfie_only_reason_mentions_liveness(self):
+        analyzer = _make_analyzer()
+        with patch.object(analyzer, "_decode_image", side_effect=_fake_decode), \
+             patch.object(analyzer, "_detect_primary_face", return_value=(50, 50, 100, 100)):
+            result = analyzer.analyze(selfie_image_base64="dGVzdA==")
+        assert "liveness" in result["reason"].lower()
+
+    def test_empty_burst_list_treated_as_selfie_only(self):
+        analyzer = _make_analyzer()
+        with patch.object(analyzer, "_decode_image", side_effect=_fake_decode), \
+             patch.object(analyzer, "_detect_primary_face", return_value=(50, 50, 100, 100)):
+            result = analyzer.analyze(
+                selfie_image_base64="dGVzdA==",
+                burst_images_base64=[],
+            )
+        assert result["is_real_person"] is False
+
+
+class TestBurstLivenessEvidence:
+    """When burst frames are provided, liveness is scored on actual signals."""
+
+    def test_burst_with_blink_and_movement_can_pass(self):
+        analyzer = _make_analyzer()
+        analyzer.config.confidence_threshold = 0.10
+        with patch.object(analyzer, "_decode_image", side_effect=_fake_decode), \
+             patch.object(analyzer, "_detect_primary_face", return_value=(50, 50, 100, 100)), \
+             patch.object(
+                 analyzer,
+                 "_analyze_burst_frames",
+                 return_value={
+                     "blink_detected": True,
+                     "head_movement_detected": True,
+                     "processed_burst_frames": 5,
+                 },
+             ):
+            result = analyzer.analyze(
+                selfie_image_base64="dGVzdA==",
+                burst_images_base64=["frame1", "frame2"],
+            )
+        assert result["checks"]["blink_detected"] is True
+        assert result["checks"]["head_movement_detected"] is True
+
+    def test_burst_without_liveness_fails(self):
+        analyzer = _make_analyzer()
+        with patch.object(analyzer, "_decode_image", side_effect=_fake_decode), \
+             patch.object(analyzer, "_detect_primary_face", return_value=(50, 50, 100, 100)), \
+             patch.object(
+                 analyzer,
+                 "_analyze_burst_frames",
+                 return_value={
+                     "blink_detected": False,
+                     "head_movement_detected": False,
+                     "processed_burst_frames": 5,
+                 },
+             ):
+            result = analyzer.analyze(
+                selfie_image_base64="dGVzdA==",
+                burst_images_base64=["frame1", "frame2"],
+            )
+        assert result["is_real_person"] is False
+        assert "liveness" in result["reason"].lower()