diff --git a/litellm/llms/replicate/chat/transformation.py b/litellm/llms/replicate/chat/transformation.py
index 4cee5489fe0..dff656db51a 100644
--- a/litellm/llms/replicate/chat/transformation.py
+++ b/litellm/llms/replicate/chat/transformation.py
@@ -7,6 +7,7 @@ from litellm.constants import REPLICATE_MODEL_NAME_WITH_ID_LENGTH
 from litellm.litellm_core_utils.prompt_templates.common_utils import (
     convert_content_list_to_str,
 )
+from litellm.litellm_core_utils.url_utils import SSRFError, assert_same_origin
 from litellm.litellm_core_utils.prompt_templates.factory import (
     custom_prompt,
     prompt_factory,
@@ -284,6 +285,10 @@ class ReplicateConfig(BaseConfig):
         ...,
         "urls":{"cancel":"https://api.replicate.com/v1/predictions/gqsmqmp1pdrj00cknr08dgmvb4/cancel","get":"https://api.replicate.com/v1/predictions/gqsmqmp1pdrj00cknr08dgmvb4","stream":"https://stream-b.svc.rno2.c.replicate.net/v1/streams/eot4gbydowuin4snhncydwxt57dfwgsc3w3snycx5nid7oef7jga"}
         }
+
+        The ``urls.get`` value is polled with the operator's Replicate token.
+        Reject off-origin URLs so a poisoned create response cannot turn the
+        proxy into credentialed SSRF (same class as Azure operation-location).
         """
         response_json: Final = response.json()
         prediction_url: Final = response_json.get("urls", {}).get("get")
@@ -293,6 +298,19 @@ class ReplicateConfig(BaseConfig):
                 message=f"LiteLLM Error - prediction url is None - {response_json}",
                 headers=response.headers,
             )
+        expected_origin = (
+            str(response.request.url)
+            if response.request is not None
+            else "https://api.replicate.com"
+        )
+        try:
+            assert_same_origin(prediction_url, expected_origin)
+        except SSRFError as ssrf_err:
+            raise ReplicateError(
+                status_code=502,
+                message=f"Rejected prediction URL: {ssrf_err}",
+                headers=response.headers,
+            )
         return prediction_url
 
     def validate_environment(
diff --git a/tests/test_litellm/llms/replicate/test_replicate_prediction_url.py b/tests/test_litellm/llms/replicate/test_replicate_prediction_url.py
new file mode 100644
index 00000000000..462e29a52c5
--- /dev/null
+++ b/tests/test_litellm/llms/replicate/test_replicate_prediction_url.py
@@ -0,0 +1,37 @@
+"""Tests for Replicate get_prediction_url same-origin guard."""
+
+import httpx
+import pytest
+
+from litellm.llms.replicate.chat.transformation import ReplicateConfig
+from litellm.llms.replicate.common_utils import ReplicateError
+
+
+def _response(urls_get: str | None, request_url: str = "https://api.replicate.com/v1/models/x/predictions"):
+    body = {"urls": {"get": urls_get}} if urls_get is not None else {"urls": {}}
+    request = httpx.Request("POST", request_url)
+    return httpx.Response(200, json=body, request=request)
+
+
+class TestReplicatePredictionUrlSameOrigin:
+    def test_accepts_same_origin_get_url(self):
+        url = ReplicateConfig().get_prediction_url(
+            _response("https://api.replicate.com/v1/predictions/abc123")
+        )
+        assert url == "https://api.replicate.com/v1/predictions/abc123"
+
+    def test_rejects_off_origin_get_url(self):
+        with pytest.raises(ReplicateError, match="Rejected prediction URL"):
+            ReplicateConfig().get_prediction_url(
+                _response("https://evil.example/steal-token")
+            )
+
+    def test_rejects_http_scheme_mismatch(self):
+        with pytest.raises(ReplicateError, match="Rejected prediction URL"):
+            ReplicateConfig().get_prediction_url(
+                _response("http://api.replicate.com/v1/predictions/abc123")
+            )
+
+    def test_missing_get_url_still_400(self):
+        with pytest.raises(ReplicateError, match="prediction url is None"):
+            ReplicateConfig().get_prediction_url(_response(None))