diff --git a/backend/app/services/ai_service.py b/backend/app/services/ai_service.py
index 3baaeea..3a119eb 100644
--- a/backend/app/services/ai_service.py
+++ b/backend/app/services/ai_service.py
@@ -216,14 +216,22 @@ def extract_questions(
     raise RuntimeError(f"Failed to extract questions after 3 attempts: {last_error}")
 
 
-def _call_model(prompt: str, model_id: str | None, api_key: str | None) -> str:
-    """Call the configured LLM and return raw text response."""
+def _call_model(prompt: str, model_id: str | None, api_key: str | None,
+                timeout: int = 180) -> str:
+    """Call the configured LLM and return raw text response.
+
+    The timeout is not optional in practice: every other call in this module has
+    one, and this one did not. A stalled connection to the proxy hung the caller
+    for good — which an interactive request survives by the user giving up, and
+    an unattended run of several hundred topics does not.
+    """
     use_model = _proxy_model(model_id or settings.LITELLM_MODEL)
     use_key = api_key or settings.LITELLM_API_KEY
     kwargs = {
         "model": use_model,
         "messages": [{"role": "user", "content": prompt}],
         "temperature": 0.1,
+        "timeout": timeout,
     }
     if use_key:
         kwargs["api_key"] = use_key
diff --git a/backend/app/services/article_writer.py b/backend/app/services/article_writer.py
index ef11fb6..a966bc6 100644
--- a/backend/app/services/article_writer.py
+++ b/backend/app/services/article_writer.py
@@ -33,6 +33,9 @@ PASSAGES = 14
 # Enough retrieved prose to write from. Below this the model is being asked to
 # write a medical article out of fragments, and it will either refuse or invent.
 MIN_SOURCE_CHARS = 3000
+# Generous for a long article, short enough that a stalled call is noticed in
+# minutes rather than discovered hours later with nothing written since.
+WRITE_TIMEOUT = 150
 # Long enough to be worth reading, short enough that nobody skims past the point.
 LONG_WORDS = 550
 SHELF = "Pediatrics"
@@ -148,7 +151,8 @@ def write_article(db: Session, topic: str, category_id: int | None = None,
                 "passages": len(passages)}
 
     model_id, api_key = get_model_for_task(db, "extraction")
-    raw = _call_model(_prompt(topic, passages), model_id, api_key)
+    # One topic must not be able to stall a run of five hundred.
+    raw = _call_model(_prompt(topic, passages), model_id, api_key, timeout=WRITE_TIMEOUT)
     text = raw.strip()
     if text.startswith("```"):
         text = re.sub(r"^```[a-z]*\n?|```$", "", text).strip()
diff --git a/backend/tests/test_ai_mode.py b/backend/tests/test_ai_mode.py
index 9a30edd..0faabae 100644
--- a/backend/tests/test_ai_mode.py
+++ b/backend/tests/test_ai_mode.py
@@ -67,6 +67,13 @@ class _AiModeBase(unittest.TestCase):
     """Fixtures shared by the route and retrieval cases; holds no tests itself."""
 
     def setUp(self):
+        # The daily limit is enforced through the real Redis, so without this a
+        # suite that passes today fails once it has been run sixty times — the
+        # tests would be measuring shared state rather than this code.
+        self._no_limit = patch("app.routers.ai_mode.check_rate_limit", lambda **kwargs: None)
+        self._no_limit.start()
+        self.addCleanup(self._no_limit.stop)
+
         self.bank = fixtures.BuilderTests()
         self.bank.setUp()
         self.client = self.bank.client