diff --git a/tests/test_gil_scoped.py b/tests/test_gil_scoped.py
index d7efc9af6b..04f7b09c71 100644
--- a/tests/test_gil_scoped.py
+++ b/tests/test_gil_scoped.py
@@ -2,6 +2,7 @@
 
 import multiprocessing
 import sys
+import sysconfig
 import threading
 import time
 
@@ -10,8 +11,11 @@
 import env
 from pybind11_tests import gil_scoped as m
 
+# Test collection seems to hold the gil
+# These tests have rare flakes in nogil; since they
+# are testing the gil, they are skipped at the moment.
 skipif_not_free_threaded = pytest.mark.skipif(
-    not getattr(sys, "_is_gil_enabled", lambda: True)(),
+    sysconfig.get_config_var("Py_GIL_DISABLED"),
     reason="Flaky without the GIL",
 )
 
diff --git a/tools/pybind11Config.cmake.in b/tools/pybind11Config.cmake.in
index fac112ecb2..f52b2fb891 100644
--- a/tools/pybind11Config.cmake.in
+++ b/tools/pybind11Config.cmake.in
@@ -94,7 +94,7 @@ you can either use the basic targets, or use the FindPython tools:
   target_link_libraries(MyModule2 PUBLIC pybind11::headers)
   set_target_properties(MyModule2 PROPERTIES
                                   INTERPROCEDURAL_OPTIMIZATION ON
-                                  CXX_VISIBILITY_PRESET ON
+                                  CXX_VISIBILITY_PRESET hidden
                                   VISIBILITY_INLINES_HIDDEN ON)
 
 If you build targets yourself, you may be interested in stripping the output