diff --git a/litellm/caching/redis_cache.py b/litellm/caching/redis_cache.py
index dd1c152a421..78e96058bae 100644
--- a/litellm/caching/redis_cache.py
+++ b/litellm/caching/redis_cache.py
@@ -391,7 +391,7 @@ class RedisCache(BaseCache):
             # Fallback for unparseable versions (e.g., "v7.0.0", "latest")
             return DEFAULT_REDIS_MAJOR_VERSION
 
-    def set_cache(self, key, value, **kwargs):
+    def set_cache(self, key, value, raise_on_error: bool = False, **kwargs):
         ttl = self.get_ttl(**kwargs)
         print_verbose(f"Set Redis Cache: key: {key}\nValue {value}\nttl={ttl}, redis_version={self.redis_version}")
         key = self.check_and_fix_namespace(key=key)
@@ -410,6 +410,8 @@ class RedisCache(BaseCache):
         except Exception as e:
             # NON blocking - notify users Redis is throwing an exception
             print_verbose(f"litellm.caching.caching: set() - Got exception from REDIS : {str(e)}")
+            if raise_on_error:
+                raise
 
     def increment_cache(self, key, value: int, ttl: Optional[float] = None, **kwargs) -> int:
         _redis_client = self.redis_client
diff --git a/litellm/proxy/management_endpoints/ui_sso.py b/litellm/proxy/management_endpoints/ui_sso.py
index 8682b61f910..088d311ca2d 100644
--- a/litellm/proxy/management_endpoints/ui_sso.py
+++ b/litellm/proxy/management_endpoints/ui_sso.py
@@ -148,6 +148,10 @@ _CLI_SSO_FLOW_CACHE_KEY_PREFIX = f"{CLI_SSO_SESSION_CACHE_KEY_PREFIX}:flow"
 _CLI_SSO_START_RATE_LIMIT_CACHE_KEY_PREFIX = f"{_CLI_SSO_FLOW_CACHE_KEY_PREFIX}:start_rate_limit"
 _CLI_SSO_START_RATE_LIMIT_WINDOW_SECONDS = 60
 _CLI_SSO_START_RATE_LIMIT_MAX_ATTEMPTS = 30
+_CLI_SSO_REDIS_UNAVAILABLE_DETAIL = (
+    "CLI login requires the proxy's configured Redis cache, which is currently unreachable. "
+    "Retry once Redis is healthy."
+)
 _CLI_SSO_USER_CODE_ALPHABET = "ABCDEFGHJKLMNPQRSTUVWXYZ23456789"
 _CLI_SSO_LOGIN_ID_RE = re.compile(r"^cli-[A-Za-z0-9_-]{12,124}$")
 _CLI_SSO_USER_CODE_RE = re.compile(rf"^[{_CLI_SSO_USER_CODE_ALPHABET}]{{4}}-[{_CLI_SSO_USER_CODE_ALPHABET}]{{4}}$")
@@ -235,11 +239,18 @@ def _check_cli_sso_start_rate_limit(
     rate_limit_cache_key = _get_cli_sso_start_rate_limit_cache_key(
         request=request, use_x_forwarded_for=use_x_forwarded_for
     )
-    current_attempts = cache.increment_cache(
-        key=rate_limit_cache_key,
-        value=1,
-        ttl=_CLI_SSO_START_RATE_LIMIT_WINDOW_SECONDS,
-    )
+    try:
+        current_attempts = cache.increment_cache(
+            key=rate_limit_cache_key,
+            value=1,
+            ttl=_CLI_SSO_START_RATE_LIMIT_WINDOW_SECONDS,
+        )
+    except Exception as e:
+        verbose_proxy_logger.error(
+            "CLI SSO start rate limit check failed because the configured Redis cache is unreachable: %s",
+            str(e),
+        )
+        raise HTTPException(status_code=503, detail=_CLI_SSO_REDIS_UNAVAILABLE_DETAIL) from e
     if current_attempts > _CLI_SSO_START_RATE_LIMIT_MAX_ATTEMPTS:
         raise HTTPException(
             status_code=429,
@@ -291,10 +302,23 @@ def _get_cli_sso_flow_or_raise(login_id: Optional[str], cache: DualCache) -> dic
 def _set_cli_sso_flow(login_id: str, cache: DualCache, flow: dict) -> None:
     cache_key = _get_cli_sso_flow_cache_key(login_id)
     redis_cache = cache.redis_cache
-    if redis_cache is not None:
-        redis_cache.set_cache(key=cache_key, value=json.dumps(flow), ttl=CLI_SSO_SESSION_TTL_SECONDS)
-    else:
+    if redis_cache is None:
         cache.set_cache(key=cache_key, value=flow, ttl=CLI_SSO_SESSION_TTL_SECONDS)
+        return
+    try:
+        redis_cache.set_cache(
+            key=cache_key,
+            value=json.dumps(flow),
+            ttl=CLI_SSO_SESSION_TTL_SECONDS,
+            raise_on_error=True,
+        )
+    except Exception as e:
+        verbose_proxy_logger.error(
+            "CLI SSO login session for login_id=%s could not be written to the configured Redis cache: %s",
+            login_id,
+            str(e),
+        )
+        raise HTTPException(status_code=503, detail=_CLI_SSO_REDIS_UNAVAILABLE_DETAIL) from e
 
 
 def _verify_cli_sso_poll_secret(flow: dict, poll_secret: Optional[str]) -> bool:
diff --git a/tests/test_litellm/caching/test_redis_cache.py b/tests/test_litellm/caching/test_redis_cache.py
index a2e18a62638..9e07bb3a844 100644
--- a/tests/test_litellm/caching/test_redis_cache.py
+++ b/tests/test_litellm/caching/test_redis_cache.py
@@ -58,6 +58,22 @@ def test_delete_cache_applies_namespace(namespace, monkeypatch, redis_no_ping):
     mock_redis_client.delete.assert_called_once_with(expected_key)
 
 
+def test_set_cache_swallows_errors_by_default_and_raises_on_opt_in(monkeypatch, redis_no_ping):
+    """Sync set_cache is fire-and-forget for callers that tolerate a cold cache,
+    but callers whose data lives only in Redis (e.g. CLI SSO login sessions)
+    must be able to opt into hearing about a failed write."""
+    monkeypatch.setenv("REDIS_HOST", "https://my-test-host")
+    redis_cache = RedisCache()
+    mock_redis_client = MagicMock()
+    mock_redis_client.set.side_effect = ConnectionError("connection refused")
+    redis_cache.redis_client = mock_redis_client
+
+    redis_cache.set_cache(key="some-key", value="some-value", ttl=60)
+
+    with pytest.raises(ConnectionError):
+        redis_cache.set_cache(key="some-key", value="some-value", ttl=60, raise_on_error=True)
+
+
 @pytest.mark.asyncio
 async def test_redis_client_init_with_socket_timeout(monkeypatch, redis_no_ping):
     monkeypatch.setenv("REDIS_HOST", "my-fake-host")
diff --git a/tests/test_litellm/proxy/management_endpoints/test_ui_sso.py b/tests/test_litellm/proxy/management_endpoints/test_ui_sso.py
index 795b7cd5a9e..6285b2d6500 100644
--- a/tests/test_litellm/proxy/management_endpoints/test_ui_sso.py
+++ b/tests/test_litellm/proxy/management_endpoints/test_ui_sso.py
@@ -2252,7 +2252,10 @@ class TestCLIKeyRegenerationFlow:
         _set_cli_sso_flow(login_id=login_id, cache=cache, flow=fresh_flow)
 
         redis_cache.set_cache.assert_called_once_with(
-            key=cache_key, value=json.dumps(fresh_flow), ttl=CLI_SSO_SESSION_TTL_SECONDS
+            key=cache_key,
+            value=json.dumps(fresh_flow),
+            ttl=CLI_SSO_SESSION_TTL_SECONDS,
+            raise_on_error=True,
         )
         cache.set_cache.assert_not_called()
 
@@ -2288,7 +2291,7 @@ class TestCLIKeyRegenerationFlow:
 
         redis_store: dict = {}
         redis_cache = MagicMock()
-        redis_cache.set_cache.side_effect = lambda key, value, ttl: redis_store.__setitem__(
+        redis_cache.set_cache.side_effect = lambda key, value, ttl, **kwargs: redis_store.__setitem__(
             key, str(value).encode("utf-8")
         )
         redis_cache.get_cache.side_effect = lambda key: RedisCache._get_cache_logic(
@@ -2363,6 +2366,58 @@ class TestCLIKeyRegenerationFlow:
         assert exc_info.value.status_code == 429
         mock_cache.set_cache.assert_not_called()
 
+    @pytest.mark.asyncio
+    async def test_cli_sso_start_returns_503_when_redis_unreachable(self):
+        """
+        Regression: when a coordination Redis is configured but unreachable, the
+        rate-limit increment used to propagate the raw ConnectionError as a 500.
+        The endpoint must fail closed with a deliberate 503 instead.
+        """
+        from litellm.proxy.management_endpoints.ui_sso import cli_sso_start
+
+        mock_request = MagicMock(spec=Request)
+        mock_request.client = SimpleNamespace(host="127.0.0.1")
+        mock_request.headers = {}
+        mock_cache = MagicMock()
+        mock_cache.increment_cache.side_effect = ConnectionError("Error 61 connecting to localhost:6379")
+
+        with (
+            patch("litellm.proxy.proxy_server.user_api_key_cache", mock_cache),
+            patch("litellm.proxy.proxy_server.cli_sso_session_cache", mock_cache),
+        ):
+            with pytest.raises(HTTPException) as exc_info:
+                await cli_sso_start(request=mock_request)
+
+        assert exc_info.value.status_code == 503
+        assert "Redis" in exc_info.value.detail
+        assert "unreachable" in exc_info.value.detail
+        mock_cache.set_cache.assert_not_called()
+
+    def test_set_cli_sso_flow_raises_503_when_redis_write_fails(self):
+        """
+        Regression: the Redis-authoritative flow write used to swallow connection
+        errors, storing the login session nowhere and deferring the failure to a
+        confusing 400 at poll time. A failed write must surface as a 503.
+        """
+        from litellm.proxy.management_endpoints.ui_sso import _set_cli_sso_flow
+
+        redis_cache = MagicMock()
+        redis_cache.set_cache.side_effect = ConnectionError("Error 61 connecting to localhost:6379")
+        cache = MagicMock()
+        cache.redis_cache = redis_cache
+
+        with pytest.raises(HTTPException) as exc_info:
+            _set_cli_sso_flow(
+                login_id="cli-redis_down_1234567890",
+                cache=cache,
+                flow={"poll_secret_hash": "hash"},
+            )
+
+        assert exc_info.value.status_code == 503
+        assert "Redis" in exc_info.value.detail
+        assert redis_cache.set_cache.call_args.kwargs["raise_on_error"] is True
+        cache.set_cache.assert_not_called()
+
     @pytest.mark.asyncio
     async def test_cli_sso_start_returns_verification_uri_complete_when_enabled(self):
         """Test CLI SSO start returns a verification_uri_complete that round-trips the user_code only when the operator opts in"""