diff --git a/CHANGELOG.md b/CHANGELOG.md
index 02d26f3c..17112fcf 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -4,6 +4,13 @@ All notable changes to this project will be documented in this file.
 
 ## [Unreleased]
 
+### Changed
+
+- OLM deployer doesn't add owner references to cluster scoped objects anymore ([#360]).
+  Owner references ensure that objects are garbage collected by OpenShift upon operator removal but they cause problems when the operator is updated.
+  This means that cluster wide objects are not removed anymore when the operator is uninstalled.
+  This behaviour is in line with the default behaviour of Helm and OLM.
+
 ## [25.11.0] - 2025-11-07
 
 ## [25.11.0-rc1] - 2025-11-06
diff --git a/rust/olm-deployer/src/owner/mod.rs b/rust/olm-deployer/src/owner/mod.rs
index 50361ebe..cf7cb4bd 100644
--- a/rust/olm-deployer/src/owner/mod.rs
+++ b/rust/olm-deployer/src/owner/mod.rs
@@ -11,9 +11,16 @@ use stackable_operator::{
     },
 };
 
-/// Updates the owner list of the `target` according to it's scope.
-/// For namespaced objects it uses the `ns_owner` whereas for cluster wide
-/// objects it uses the `cluster_owner`.
+/// Updates owner references of objects created by this deployer so that when an operator is
+/// uninstalled by OLM, all created objects are also removed by Kubernetes garbage collection.
+///
+/// Namespaced object's owner references are updated in place with the value of `ns_owner`.
+///
+/// A previous version of this function also updated cluster scoped objects to set the owner
+/// reference to `cluster_owner`, but this turned out to be problematic. First, this is not how OLM
+/// and Helm behave by default. Second, when updating the listener operator, deleting and recreating
+/// the CSI driver definition the cluster was left in a broken state.
+/// See: this comment for more details: https://github.com/stackabletech/issues/issues/799#issuecomment-3601121617
 pub(super) fn maybe_update_owner(
     target: &mut DynamicObject,
     scope: &Scope,
@@ -21,9 +28,13 @@ pub(super) fn maybe_update_owner(
     cluster_owner: &ClusterRole,
 ) -> Result<()> {
     let owner_ref = owner_ref(scope, ns_owner, cluster_owner)?;
-    match target.metadata.owner_references {
-        Some(ref mut ors) => ors.push(owner_ref),
-        None => target.metadata.owner_references = Some(vec![owner_ref]),
+    // 2025-12-10: do not set owner references for cluster scoped objects anymore to prevent them from being
+    // deleted upon operator upgrades.
+    if scope == &Scope::Namespaced {
+        match target.metadata.owner_references {
+            Some(ref mut ors) => ors.push(owner_ref),
+            None => target.metadata.owner_references = Some(vec![owner_ref]),
+        }
     }
     Ok(())
 }
@@ -147,13 +158,7 @@ rules:
         let mut daemonset = DAEMONSET.clone();
         maybe_update_owner(&mut daemonset, &Scope::Cluster, &DEPLOYMENT, &CLUSTER_ROLE)?;
 
-        let expected = Some(vec![OwnerReference {
-            uid: "d9287d0a-3069-47c3-8c90-b714dc6dddaa".to_string(),
-            name: "listener-operator-clusterrole".to_string(),
-            kind: "ClusterRole".to_string(),
-            api_version: "rbac.authorization.k8s.io/v1".to_string(),
-            ..OwnerReference::default()
-        }]);
+        let expected = None;
         assert_eq!(daemonset.metadata.owner_references, expected);
         Ok(())
     }