diff --git a/tensorrt_llm/_torch/kv_cache_compression/triattention/triattention.py b/tensorrt_llm/_torch/kv_cache_compression/triattention/triattention.py
index de0e1ef4f04a..04ad73b627df 100644
--- a/tensorrt_llm/_torch/kv_cache_compression/triattention/triattention.py
+++ b/tensorrt_llm/_torch/kv_cache_compression/triattention/triattention.py
@@ -277,7 +277,7 @@ def _resolve_attention_layers(self) -> Tuple[List[int], List[int], Optional[int]
         return (dense_layers, swa_layers, window_size)
 
     def _load_calibration(self) -> None:
-        raw = torch.load(self.calibration_path, map_location="cpu", weights_only=False)
+        raw = torch.load(self.calibration_path, map_location="cpu", weights_only=True)
         if isinstance(raw, dict) and _REQUIRED_CALIBRATION_KEYS <= set(raw):
             e_q = raw["E_q"]
             e_q_norm = raw["E_q_norm"]
diff --git a/tensorrt_llm/_torch/pyexecutor/profiling.py b/tensorrt_llm/_torch/pyexecutor/profiling.py
index e7bcbd5157fd..8cbb20a9d664 100644
--- a/tensorrt_llm/_torch/pyexecutor/profiling.py
+++ b/tensorrt_llm/_torch/pyexecutor/profiling.py
@@ -43,6 +43,7 @@
 
 import functools
 import os
+import tempfile
 import time
 import uuid
 import weakref
@@ -287,15 +288,7 @@ def _prepare_and_broadcast_start(
             activities = ["CPU", "GPU"]
 
         if output_dir is None:
-            # ``/tmp`` is the documented developer default for the
-            # profile trace directory (matches the server docs); the
-            # user-facing override is the
-            # ``TLLM_TORCH_PROFILER_DIR`` env var or the
-            # ``output_dir`` request body field.
-            output_dir = os.environ.get(
-                "TLLM_TORCH_PROFILER_DIR",
-                "/tmp",  # nosec B108
-            )
+            output_dir = os.environ.get("TLLM_TORCH_PROFILER_DIR")
 
         # Fail fast if the directory cannot be created. Returning
         # success here would schedule a profile window that later
@@ -306,11 +299,16 @@ def _prepare_and_broadcast_start(
         # plumbing as the "already in progress" guard above, so the
         # HTTP handler can map it to a 5xx with a meaningful body.
         try:
-            os.makedirs(output_dir, exist_ok=True)
+            if output_dir is None:
+                with tempfile.TemporaryDirectory(prefix="trtllm-profile-"):
+                    pass
+            else:
+                os.makedirs(output_dir, exist_ok=True)
         except OSError as e:
             from tensorrt_llm.executor.utils import RequestError
 
-            raise RequestError(f"Failed to create profile output_dir {output_dir}: {e}") from e
+            target = output_dir if output_dir is not None else "the system temporary directory"
+            raise RequestError(f"Failed to create profile output directory at {target}: {e}") from e
 
         # Include a unique profile id (monotonic timestamp + short
         # uuid fragment) so successive start/stop cycles under the
@@ -453,9 +451,11 @@ def apply_start_config(self, config: Dict) -> None:
         """
         executor = self._executor
         activities = config.get("activities") or ["CPU", "GPU"]
-        # ``/tmp`` matches the documented developer default; see
-        # ``start_profile`` for the rationale.
-        output_dir = config.get("output_dir") or "/tmp"  # nosec B108
+        output_dir = config.get("output_dir")
+        if not output_dir:
+            # Maintain compatibility for callers constructing
+            # configs without output_dir.
+            output_dir = tempfile.mkdtemp(prefix="trtllm-profile-")
         profile_id = config.get("profile_id") or f"{int(time.time())}-{uuid.uuid4().hex[:8]}"
         # Prefer the legacy absolute ``start_iter`` form (older unit
         # tests and env-var driven paths set this directly). For the
diff --git a/tensorrt_llm/_torch/pyexecutor/py_executor.py b/tensorrt_llm/_torch/pyexecutor/py_executor.py
index 44768cbdb1bf..4ae5999c510c 100644
--- a/tensorrt_llm/_torch/pyexecutor/py_executor.py
+++ b/tensorrt_llm/_torch/pyexecutor/py_executor.py
@@ -2070,7 +2070,7 @@ def start_profile(self,
         Args:
             output_dir: Directory where chrome traces are written. If
                 ``None``, falls back to the ``TLLM_TORCH_PROFILER_DIR``
-                environment variable and finally to ``/tmp``.
+                environment variable and finally to a temporary directory.
             num_steps: Number of iterations to profile. If ``None``,
                 profiling runs until ``stop_profile`` is called.
             start_step: Additional iterations to skip before profiling
diff --git a/tensorrt_llm/commands/serve.py b/tensorrt_llm/commands/serve.py
index d21e1b6303cc..9aedeeb874fa 100644
--- a/tensorrt_llm/commands/serve.py
+++ b/tensorrt_llm/commands/serve.py
@@ -410,9 +410,8 @@ def _publish_bound_address(report_addr: Optional[str], host: str,
         host = resolved
     report_addr = os.path.abspath(report_addr)
     parent = os.path.dirname(report_addr)
-    if parent:
-        os.makedirs(parent, exist_ok=True)
-    fd, tmp_path = tempfile.mkstemp(dir=parent or None,
+    os.makedirs(parent, exist_ok=True)
+    fd, tmp_path = tempfile.mkstemp(dir=parent,
                                     prefix=os.path.basename(report_addr) + ".",
                                     suffix=".tmp")
     try:
@@ -2253,9 +2252,18 @@ def _serve_coordinator_and_fleet(disagg_cfg, config_file,
     # that dominated per-request latency). Opt-out via the UDS env = "0".
     use_uds = os.environ.get(DisaggWorkerEnvs.TLLM_DISAGG_COORDINATOR_UDS,
                              "1") == "1"
-    coord_uds = (
-        f"/tmp/trtllm_disagg_coord_{public_port}.sock"  # nosec B108
-        if use_uds else None)
+    coord_uds_dir = (tempfile.TemporaryDirectory(
+        prefix="trtllm-disagg-coord-") if use_uds else None)
+    coord_uds = (os.path.join(coord_uds_dir.name, "coordinator.sock")
+                 if coord_uds_dir else None)
+    # TMPDIR must be at most 61 bytes to keep this socket path within Linux's
+    # 107-byte limit.
+    if (coord_uds_dir is not None and coord_uds is not None
+            and len(os.fsencode(coord_uds)) > 107):
+        coord_uds_dir.cleanup()
+        raise ValueError(
+            "Coordinator Unix socket path exceeds 107 bytes; TMPDIR must be "
+            "at most 61 bytes")
     # Fleet points at the UDS when enabled; the TCP port stays up for health.
     coord_url = f"unix:{coord_uds}" if coord_uds else \
         f"http://{public_host}:{coord_port}"
@@ -2343,6 +2351,8 @@ async def _monitor_fleet():
         for process in fleet:
             if process.poll() is None:
                 process.terminate()
+        if coord_uds_dir is not None:
+            coord_uds_dir.cleanup()
 
 
 def _init_fleet_worker_process():
diff --git a/tensorrt_llm/llmapi/llm.py b/tensorrt_llm/llmapi/llm.py
index fdda58cbbf31..64698cf5d44d 100644
--- a/tensorrt_llm/llmapi/llm.py
+++ b/tensorrt_llm/llmapi/llm.py
@@ -1671,7 +1671,7 @@ def start_profile(self,
             output_dir (str, optional): Directory where chrome traces are
                 written. If ``None``, falls back to the
                 ``TLLM_TORCH_PROFILER_DIR`` environment variable and finally
-                to ``/tmp``. Defaults to None.
+                to a temporary directory. Defaults to None.
             num_steps (int, optional): Number of engine iterations to
                 capture. When set, the engine stops the window automatically
                 and ``stop_profile`` does not need to be called. When
diff --git a/tensorrt_llm/serve/openai_protocol.py b/tensorrt_llm/serve/openai_protocol.py
index 0d85fd554229..31cc276a1adb 100644
--- a/tensorrt_llm/serve/openai_protocol.py
+++ b/tensorrt_llm/serve/openai_protocol.py
@@ -2375,7 +2375,7 @@ class StartProfileRequest(OpenAIBaseModel):
         default=None,
         description="Directory where chrome traces are written. Defaults "
         "to the TLLM_TORCH_PROFILER_DIR environment variable, "
-        "then /tmp.")
+        "then a temporary directory.")
     num_steps: Optional[PositiveInt] = Field(
         default=None,
         description="Number of iterations to profile. Must be >= 1 if "