diff --git a/CHANGELOG.md b/CHANGELOG.md
index 483c73d1d..fd2c78b91 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -82,6 +82,8 @@ to include examples, links to docs, or any other relevant information.
 
 ### Fixed
 
+- Continue-as-new requests from workflow update handlers now fail the workflow
+  task instead of leaving the update unresolved.
 - Marked system Nexus envelope payloads so nested payloads can be detected and
   visited after the envelope is already stored as a payload.
 
diff --git a/temporalio/worker/_workflow_instance.py b/temporalio/worker/_workflow_instance.py
index d0b10ccae..9d956aac4 100644
--- a/temporalio/worker/_workflow_instance.py
+++ b/temporalio/worker/_workflow_instance.py
@@ -761,6 +761,14 @@ async def run_update() -> None:
                 else:
                     self._current_activation_error = err
                     return
+            except _ContinueAsNewError:
+                # Continue-as-new is only valid from the top-level workflow.
+                # From an update handler it must fail the workflow task
+                # instead of escaping this detached task.
+                self._current_activation_error = RuntimeError(
+                    "Cannot continue as new from an update handler"
+                )
+                return
             except BaseException:
                 if self._deleting:
                     if LOG_IGNORE_DURING_DELETE:
diff --git a/tests/worker/test_workflow.py b/tests/worker/test_workflow.py
index 283357c2d..68861bada 100644
--- a/tests/worker/test_workflow.py
+++ b/tests/worker/test_workflow.py
@@ -5259,6 +5259,7 @@ async def do_stuff(buffer: MetricBuffer) -> None:
 
 bad_validator_fail_ct = 0
 task_fail_ct = 0
+continue_as_new_update_task_fail_ct = 0
 
 
 @workflow.defn
@@ -5342,6 +5343,14 @@ def throws_runtime_err(self) -> None:
             task_fail_ct += 1
             raise RuntimeError("intentional failure")
 
+    @workflow.update
+    def continue_as_new_for_task_failure(self) -> str:
+        global continue_as_new_update_task_fail_ct
+        if continue_as_new_update_task_fail_ct < 1:
+            continue_as_new_update_task_fail_ct += 1
+            workflow.continue_as_new()
+        return "recovered after task failure"
+
 
 async def test_workflow_update_handlers_happy(client: Client):
     async with new_worker(
@@ -5484,6 +5493,31 @@ async def test_workflow_update_task_fails(client: Client):
         assert bad_validator_fail_ct == 2
 
 
+async def test_workflow_update_continue_as_new_fails_task(client: Client):
+    # Continue-as-new is a control-flow BaseException. From an update handler
+    # it must fail the workflow task, then retry the update rather than letting
+    # the detached update task escape without completing the protocol message.
+    global continue_as_new_update_task_fail_ct
+    continue_as_new_update_task_fail_ct = 0
+    async with new_worker(
+        client, UpdateHandlersWorkflow, workflow_runner=UnsandboxedWorkflowRunner()
+    ) as worker:
+        handle = await client.start_workflow(
+            UpdateHandlersWorkflow.run,
+            id=f"update-continue-as-new-{uuid.uuid4()}",
+            task_queue=worker.task_queue,
+            task_timeout=timedelta(seconds=1),
+        )
+        result = await asyncio.wait_for(
+            handle.execute_update(
+                UpdateHandlersWorkflow.continue_as_new_for_task_failure
+            ),
+            timeout=timedelta(seconds=10).total_seconds(),
+        )
+        assert result == "recovered after task failure"
+        assert continue_as_new_update_task_fail_ct == 1
+
+
 @workflow.defn
 class UpdateRespectsFirstExecutionRunIdWorkflow:
     def __init__(self) -> None: