diff --git a/crates/proxy/src/runtime.rs b/crates/proxy/src/runtime.rs
index 4b84b59..5d9ee33 100644
--- a/crates/proxy/src/runtime.rs
+++ b/crates/proxy/src/runtime.rs
@@ -270,6 +270,9 @@ impl ChatCompletionRuntime {
                     "all deployments for this model are at their RPM limit".to_string(),
                 ));
             }
+            return Err(ChatCompletionError::InvalidRequest(format!(
+                "model '{model}' is not configured in model_list"
+            )));
         }
 
         let state = self
diff --git a/crates/proxy/src/server/bedrock_passthrough.rs b/crates/proxy/src/server/bedrock_passthrough.rs
index 3dbb5cc..0c23849 100644
--- a/crates/proxy/src/server/bedrock_passthrough.rs
+++ b/crates/proxy/src/server/bedrock_passthrough.rs
@@ -96,6 +96,14 @@ pub(crate) async fn bedrock_passthrough(
             );
             return (StatusCode::TOO_MANY_REQUESTS, Json(err)).into_response();
         }
+        super::state::ResolvedModel::UnknownModel => {
+            let err = anyllm_translate::mapping::errors_map::create_anthropic_error(
+                anyllm_translate::anthropic::ErrorType::InvalidRequestError,
+                format!("model '{}' is not configured in model_list", model_id),
+                None,
+            );
+            return (StatusCode::BAD_REQUEST, Json(err)).into_response();
+        }
         super::state::ResolvedModel::Legacy(m) => m,
     };
 
diff --git a/crates/proxy/src/server/state.rs b/crates/proxy/src/server/state.rs
index 430272e..480359b 100644
--- a/crates/proxy/src/server/state.rs
+++ b/crates/proxy/src/server/state.rs
@@ -52,6 +52,8 @@ pub(crate) enum ResolvedModel {
     },
     /// Model is known but all deployments are at their RPM limit.
     AllAtLimit,
+    /// Model router is active but the model alias is not configured.
+    UnknownModel,
     /// No model router, or model not in router. Used legacy ModelMapping.
     Legacy(String),
 }
@@ -140,6 +142,7 @@ impl AppState {
             if router.has_model(model) {
                 return ResolvedModel::AllAtLimit;
             }
+            return ResolvedModel::UnknownModel;
         }
         ResolvedModel::Legacy(self.map_model(model))
     }
@@ -200,6 +203,14 @@ impl AppState {
                 );
                 Err((StatusCode::TOO_MANY_REQUESTS, Json(err)).into_response())
             }
+            ResolvedModel::UnknownModel => {
+                let err = mapping::errors_map::create_anthropic_error(
+                    anthropic::ErrorType::InvalidRequestError,
+                    format!("model '{model}' is not configured in model_list"),
+                    None,
+                );
+                Err((StatusCode::BAD_REQUEST, Json(err)).into_response())
+            }
             ResolvedModel::Legacy(mapped) => Ok((mapped, self.clone(), None)),
         }
     }
diff --git a/crates/proxy/tests/runtime.rs b/crates/proxy/tests/runtime.rs
index 655a0b8..35c27c7 100644
--- a/crates/proxy/tests/runtime.rs
+++ b/crates/proxy/tests/runtime.rs
@@ -198,6 +198,42 @@ async fn runtime_model_router_selects_backend_and_mapped_model() {
     );
 }
 
+#[tokio::test]
+async fn runtime_model_router_rejects_unconfigured_model() {
+    let captured_a = Arc::new(Mutex::new(None));
+    let base_a = spawn_json_backend(captured_a.clone(), chat_response("unused")).await;
+
+    let mut backends = IndexMap::new();
+    backends.insert("a".to_string(), backend_config(BackendKind::OpenAI, base_a));
+
+    let deployment = Arc::new(anyllm_proxy::config::model_router::Deployment::new(
+        "a".to_string(),
+        "actual-model".to_string(),
+        None,
+        None,
+    ));
+    let mut routes = HashMap::new();
+    routes.insert("virtual-model".to_string(), vec![deployment]);
+    let router = Arc::new(RwLock::new(
+        anyllm_proxy::config::model_router::ModelRouter::new(routes),
+    ));
+
+    let runtime = ChatCompletionRuntime::from_multi_config_with_model_router(
+        multi_config(backends, "a"),
+        Some(router),
+    );
+    let req = serde_json::from_value(json!({
+        "model": "unconfigured-model",
+        "messages": [{"role": "user", "content": "hello"}],
+        "max_tokens": 32
+    }))
+    .unwrap();
+
+    let err = runtime.complete(req).await.unwrap_err();
+    assert!(matches!(err, ChatCompletionError::InvalidRequest(_)));
+    assert!(captured_a.lock().unwrap().is_none());
+}
+
 #[tokio::test]
 async fn runtime_returns_typed_error_for_unsupported_backend() {
     let mut backends = IndexMap::new();