diff --git a/tests/integration/kubernetes/k8s-plain-ephemeral-data-storage.bats b/tests/integration/kubernetes/k8s-plain-ephemeral-data-storage.bats
new file mode 100644
index 0000000000..1177df5ed8
--- /dev/null
+++ b/tests/integration/kubernetes/k8s-plain-ephemeral-data-storage.bats
@@ -0,0 +1,304 @@
+#!/usr/bin/env bats
+#
+# Copyright (c) 2026 NVIDIA Corporation
+#
+# SPDX-License-Identifier: Apache-2.0
+#
+
+load "${BATS_TEST_DIRNAME}/lib.sh"
+load "${BATS_TEST_DIRNAME}/../../common.bash"
+load "${BATS_TEST_DIRNAME}/tests_common.sh"
+
+export KATA_HYPERVISOR="${KATA_HYPERVISOR:-qemu}"
+
+skip_unsupported_runtime() {
+	# The runtime-rs block emptyDir path uses BlockModern, inherited from the
+	# older block-encrypted implementation. The encrypted test is CoCo-only, so
+	# it never reached clh-runtime-rs or dragonball in CI; this generic
+	# block-plain test would, so skip those unsupported VMM paths explicitly.
+	case "${KATA_HYPERVISOR}" in
+		clh-runtime-rs|clh-azure-runtime-rs|dragonball)
+			skip "block-plain emptyDir uses runtime-rs BlockModern, whose VMM glue is missing for ${KATA_HYPERVISOR}"
+			;;
+		clh-azure)
+			skip "block-plain emptyDir requires guest mkfs.ext4, which is not included in the Azure Linux UVM rootfs"
+			;;
+	esac
+}
+
+# Return the full mountinfo row whose field 5 is the tested mount point.
+mountinfo_for_mountpoint() {
+	pod_exec "${pod_name}" sh -c \
+		"awk -v mp='${mountpoint}' '\$5 == mp {print; exit}' /proc/self/mountinfo"
+}
+
+# Return a field after the "-" separator: 1 = fs type, 2 = source, 3 = super options.
+mountinfo_after_separator_field() {
+	local field_offset="$1"
+
+	pod_exec "${pod_name}" sh -c \
+		"awk -v mp='${mountpoint}' '\$5 == mp {for (i = 1; i <= NF; i++) if (\$i == \"-\") {print \$(i + ${field_offset}); exit}}' /proc/self/mountinfo"
+}
+
+# Return field 6 mount options plus super options, which are three fields after "-".
+mount_options() {
+	pod_exec "${pod_name}" sh -c \
+		"awk -v mp='${mountpoint}' '\$5 == mp {for (i = 1; i <= NF; i++) if (\$i == \"-\") {print \$6 \",\" \$(i + 3); exit}}' /proc/self/mountinfo"
+}
+
+# Return the allocated host bytes for a disk image, not its apparent size.
+host_disk_allocated_bytes() {
+	local disk_path="$1"
+	exec_host "${node}" "du -B1 '${disk_path}' | awk 'NR == 1 {print \$1}'"
+}
+
+host_disk_path() {
+	local pod_uid
+	pod_uid="$(kubectl get pod "${pod_name}" -o jsonpath='{.metadata.uid}')"
+	echo "$(get_kubelet_data_dir)/pods/${pod_uid}/volumes/kubernetes.io~empty-dir/${volume_name}/disk.img"
+}
+
+pod_evicted() {
+	kubectl get pod "${pod_name}" -o jsonpath='{.status.reason}' 2>/dev/null | grep -q '^Evicted$'
+}
+
+pod_eviction_message() {
+	kubectl get pod "${pod_name}" -o jsonpath='{.status.message}' 2>/dev/null || true
+}
+
+pod_events() {
+	kubectl get events \
+		--field-selector "involvedObject.kind=Pod,involvedObject.name=${pod_name}" \
+		-o jsonpath='{range .items[*]}{.reason}{" "}{.message}{"\n"}{end}' 2>/dev/null || true
+}
+
+size_limit_eviction_observed() {
+	local events
+	local message
+	local size_limit_pattern="(exceed|exceeded|exceeds).*(emptyDir|ephemeral|limit|sizeLimit|${volume_name})"
+
+	if pod_evicted; then
+		return 0
+	fi
+
+	message="$(pod_eviction_message)"
+	if echo "${message}" | grep -Eqi "${size_limit_pattern}"; then
+		return 0
+	fi
+
+	events="$(pod_events)"
+	echo "${events}" | grep -Eqi "evict" || return 1
+	echo "${events}" | grep -Eqi "${size_limit_pattern}"
+}
+
+host_disk_exceeds_allocated_bytes_or_pod_evicted() {
+	local allocated_bytes
+	local disk_path="$1"
+	local min_allocated_bytes="$2"
+
+	if pod_evicted; then
+		info "pod ${pod_name} is already evicted while waiting for sizeLimit accounting: $(pod_eviction_message)"
+		return 0
+	fi
+
+	if ! exec_host "${node}" "test -f '${disk_path}'" >/dev/null 2>&1; then
+		info "host disk image is not present while waiting for sizeLimit accounting: ${disk_path}"
+		if pod_evicted; then
+			info "pod ${pod_name} was evicted after host disk image disappeared: $(pod_eviction_message)"
+			return 0
+		fi
+
+		pod_evicted
+		return $?
+	fi
+
+	allocated_bytes="$(host_disk_allocated_bytes "${disk_path}" 2>/dev/null || echo 0)"
+	info "host disk allocated bytes while waiting for sizeLimit accounting: ${allocated_bytes}"
+	[[ "${allocated_bytes}" =~ ^[0-9]+$ ]] || return 1
+	(( allocated_bytes > min_allocated_bytes )) || pod_evicted
+}
+
+set_emptydir_size_limit() {
+	local size_limit="$1"
+
+	yq -i "(.spec.volumes[] | select(.name == \"${volume_name}\").emptyDir.sizeLimit) = \"${size_limit}\"" "${yaml_file}"
+}
+
+apply_plain_ephemeral_pod() {
+	auto_generate_policy "${policy_settings_dir}" "${yaml_file}"
+	kubectl apply -f "${yaml_file}"
+}
+
+guest_discard_max_bytes() {
+	pod_exec "${pod_name}" sh -c \
+		"src=\$(awk -v mp='${mountpoint}' '\$5 == mp {for (i = 1; i <= NF; i++) if (\$i == \"-\") {print \$(i + 2); exit}}' /proc/self/mountinfo); \
+		blk=\${src#/dev/}; blk=\${blk##*/}; \
+		if [ -e \"/sys/class/block/\${blk}/partition\" ]; then blk=\$(basename \"\$(readlink -f \"/sys/class/block/\${blk}/..\")\"); fi; \
+		cat \"/sys/class/block/\${blk}/queue/discard_max_bytes\" 2>/dev/null || echo 0"
+}
+
+setup() {
+	local runtime_config_dropin_file
+
+	skip_unsupported_runtime
+	setup_common || die "setup_common failed"
+
+	pod_name="plain-ephemeral-data-storage"
+	volume_name="temp-plain"
+	mountpoint="/mnt/temp-plain"
+	yaml_template="${pod_config_dir}/pod-plain-ephemeral-data-storage.yaml.in"
+	yaml_file="${pod_config_dir}/pod-plain-ephemeral-data-storage.yaml"
+
+	RUNTIMECLASS="kata-${KATA_HYPERVISOR}" envsubst "\${RUNTIMECLASS}" \
+		< "${yaml_template}" > "${yaml_file}"
+
+	runtime_config_dropin_file="${BATS_FILE_TMPDIR}/99-k8s-plain-ephemeral-data-storage.toml"
+	cat > "${runtime_config_dropin_file}" <<EOF
+[runtime]
+emptydir_mode = "block-plain"
+EOF
+	runtime_config_dropin="$(set_kata_runtime_config_dropin_file \
+		"${node}" \
+		"${runtime_config_dropin_file}")" || \
+		skip "No Kata runtime config found for ${KATA_HYPERVISOR}"
+
+	if [[ "${PULL_TYPE:-default}" == "guest-pull" ]] && is_confidential_runtime_class "${KATA_HYPERVISOR}"; then
+		set_metadata_annotation "${yaml_file}" \
+			"io.containerd.cri.runtime-handler" \
+			"kata-${KATA_HYPERVISOR}"
+	fi
+
+	policy_settings_dir="$(create_tmp_policy_settings_dir "${pod_config_dir}")"
+	set_genpolicy_emptydir_type "${policy_settings_dir}" "block-plain"
+	allow_requests "${policy_settings_dir}" "ExecProcessRequest" "ReadStreamRequest"
+}
+
+@test "Plain ephemeral data storage" {
+	apply_plain_ephemeral_pod
+	kubectl wait --for=condition=Ready --timeout="${timeout}" pod "${pod_name}"
+
+	# Verify that emptyDir is mounted as a plain ext4 block device.
+	mountinfo="$(mountinfo_for_mountpoint)"
+	info "mountinfo for ${mountpoint}: ${mountinfo}"
+	[[ -n "${mountinfo}" ]]
+
+	fs_type="$(mountinfo_after_separator_field 1)"
+	source="$(mountinfo_after_separator_field 2)"
+	info "mount source: ${source}"
+	info "mount filesystem type: ${fs_type}"
+
+	[[ "${fs_type}" == "ext4" ]]
+	[[ "${source}" == /dev/* ]]
+	[[ "${source}" != /dev/mapper/* ]]
+	[[ "${source}" != "tmpfs" ]]
+
+	[[ "$(pod_exec "${pod_name}" sh -c "echo foo > '${mountpoint}/foo.txt' && cat '${mountpoint}/foo.txt'")" == "foo" ]]
+	pod_exec "${pod_name}" sh -c "dd if=/dev/zero of='${mountpoint}/blob' bs=1M count=32 && sync"
+}
+
+@test "Plain ephemeral data storage sizeLimit evicts pod" {
+	local accounting_wait_time="${wait_time}"
+	local disk_path
+	local eviction_wait_time=120
+	local events
+	local size_limit="512Mi"
+	local size_limit_bytes=$((512 * 1024 * 1024))
+	local write_mib=2048
+
+	set_emptydir_size_limit "${size_limit}"
+	apply_plain_ephemeral_pod
+	kubectl wait --for=condition=Ready --timeout="${timeout}" pod "${pod_name}"
+
+	disk_path="$(host_disk_path)"
+
+	pod_exec "${pod_name}" sh -c "dd if=/dev/zero of='${mountpoint}/size-limit-test.bin' bs=1M count=${write_mib} conv=fsync" || true
+
+	# First wait for proof that the oversized write crossed the emptyDir limit:
+	# either the host disk image exceeds the limit, or kubelet already evicted the
+	# pod and may have removed disk.img. This distinguishes "we did not write too
+	# much" from "we wrote too much, but eviction evidence did not appear".
+	waitForProcess "${accounting_wait_time}" "${sleep_time}" \
+		"host_disk_exceeds_allocated_bytes_or_pod_evicted '${disk_path}' '${size_limit_bytes}'"
+
+	# Eviction has been observed to take slightly more than the generic 90s wait.
+	waitForProcess "${eviction_wait_time}" "${sleep_time}" size_limit_eviction_observed
+	info "pod ${pod_name} reason: $(kubectl get pod "${pod_name}" -o jsonpath='{.status.reason}' 2>/dev/null || true)"
+	info "pod ${pod_name} eviction message: $(pod_eviction_message)"
+
+	events="$(pod_events)"
+	info "events for ${pod_name}: ${events}"
+}
+
+@test "Plain ephemeral data storage discard reclaims host blocks" {
+	local allocated_after_delete
+	local allocated_after_write
+	local allocated_before
+	local apparent_size
+	local discard_max
+	local disk_path
+	local min_delta
+	local options
+
+	apply_plain_ephemeral_pod
+	kubectl wait --for=condition=Ready --timeout="${timeout}" pod "${pod_name}"
+
+	# Reclaim requires discard support in the guest and discard on the mount.
+	discard_max="$(guest_discard_max_bytes | tr -d '\r\n')"
+	[[ "${discard_max}" =~ ^[0-9]+$ ]] || discard_max=0
+	(( discard_max > 0 )) || skip "Block device for ${mountpoint} does not advertise discard support"
+
+	options="$(mount_options)"
+	info "mount options for ${mountpoint}: ${options}"
+	[[ ",${options}," == *",discard,"* ]]
+
+	# Track the sparse host disk image backing this emptyDir volume.
+	disk_path="$(host_disk_path)"
+	exec_host "${node}" "test -f '${disk_path}'" || die "Missing host disk image ${disk_path}"
+
+	apparent_size="$(exec_host "${node}" "stat -c %s '${disk_path}'")"
+	allocated_before="$(host_disk_allocated_bytes "${disk_path}")"
+	info "host disk apparent size before write: ${apparent_size}"
+	info "host disk allocated bytes before write: ${allocated_before}"
+	(( allocated_before < apparent_size ))
+
+	# Write enough data to force host allocation growth.
+	pod_exec "${pod_name}" sh -c "dd if=/dev/zero of='${mountpoint}/discard-test.bin' bs=1M count=96 conv=fsync"
+	exec_host "${node}" sync
+
+	allocated_after_write="$(host_disk_allocated_bytes "${disk_path}")"
+	info "host disk allocated bytes after write: ${allocated_after_write}"
+
+	min_delta=$((32 * 1024 * 1024))
+	(( allocated_after_write >= allocated_before + min_delta ))
+
+	# Removing the guest file should discard blocks from the host image.
+	wait_for_host_disk_reclaim() {
+		allocated_after_delete="$(host_disk_allocated_bytes "${disk_path}")"
+		info "host disk allocated bytes after delete: ${allocated_after_delete}"
+		(( allocated_after_delete <= allocated_after_write - min_delta ))
+	}
+
+	pod_exec "${pod_name}" sh -c "rm -f '${mountpoint}/discard-test.bin' && sync"
+	exec_host "${node}" sync
+
+	waitForProcess "${wait_time}" "${sleep_time}" wait_for_host_disk_reclaim
+}
+
+teardown() {
+	skip_unsupported_runtime
+
+	echo "=== Plain ephemeral data storage pod describe ==="
+	kubectl describe pod "${pod_name:-plain-ephemeral-data-storage}" || true
+
+	echo "=== Plain ephemeral data storage pod logs ==="
+	kubectl logs "${pod_name:-plain-ephemeral-data-storage}" || true
+
+	# Always restore the Kata config (no-op if no drop-in was applied).
+	remove_kata_runtime_config_dropin_file "${node}" "${runtime_config_dropin:-}" || true
+
+	delete_tmp_policy_settings_dir "${policy_settings_dir:-}"
+
+	[[ -f "${yaml_file:-}" ]] && kubectl delete -f "${yaml_file}" --ignore-not-found=true
+
+	print_node_journal_since_test_start "${node}" "${node_start_time:-}" "${BATS_TEST_COMPLETED:-}"
+}
diff --git a/tests/integration/kubernetes/k8s-trusted-ephemeral-data-storage.bats b/tests/integration/kubernetes/k8s-trusted-ephemeral-data-storage.bats
index 42aaa152b3..a19f7ed42d 100644
--- a/tests/integration/kubernetes/k8s-trusted-ephemeral-data-storage.bats
+++ b/tests/integration/kubernetes/k8s-trusted-ephemeral-data-storage.bats
@@ -23,6 +23,7 @@ setup() {
 
     yaml_file="${pod_config_dir}/pod-trusted-ephemeral-data-storage.yaml"
     policy_settings_dir="$(create_tmp_policy_settings_dir "${pod_config_dir}")"
+    set_genpolicy_emptydir_type "${policy_settings_dir}" "block-encrypted"
 
     # The policy would only block container creation, so allow these
     # requests to make writing tests easier.
diff --git a/tests/integration/kubernetes/run_kubernetes_tests.sh b/tests/integration/kubernetes/run_kubernetes_tests.sh
index 511ca0d5b2..9fb3bf3d3b 100755
--- a/tests/integration/kubernetes/run_kubernetes_tests.sh
+++ b/tests/integration/kubernetes/run_kubernetes_tests.sh
@@ -94,6 +94,7 @@ else
 		"k8s-optional-empty-configmap.bats" \
 		"k8s-optional-empty-secret.bats" \
 		"k8s-pid-ns.bats" \
+		"k8s-plain-ephemeral-data-storage.bats" \
 		"k8s-pod-quota.bats" \
 		"k8s-port-forward.bats" \
 		"k8s-privileged.bats" \
diff --git a/tests/integration/kubernetes/runtimeclass_workloads/pod-plain-ephemeral-data-storage.yaml.in b/tests/integration/kubernetes/runtimeclass_workloads/pod-plain-ephemeral-data-storage.yaml.in
new file mode 100644
index 0000000000..f90ead81cf
--- /dev/null
+++ b/tests/integration/kubernetes/runtimeclass_workloads/pod-plain-ephemeral-data-storage.yaml.in
@@ -0,0 +1,27 @@
+#
+# Copyright (c) 2026 NVIDIA Corporation
+#
+# SPDX-License-Identifier: Apache-2.0
+#
+kind: Pod
+apiVersion: v1
+metadata:
+  name: plain-ephemeral-data-storage
+spec:
+  securityContext:
+    runAsUser: 0
+    runAsGroup: 0
+    supplementalGroups: [1, 2, 3, 4, 6, 10, 11, 20, 26, 27]
+  runtimeClassName: ${RUNTIMECLASS}
+  restartPolicy: Never
+  containers:
+  - image: quay.io/prometheus/busybox:latest
+    name: test
+    command: ["sleep", "infinity"]
+    volumeMounts:
+    - name: temp-plain
+      mountPath: /mnt/temp-plain
+  volumes:
+  - name: temp-plain
+    emptyDir:
+      sizeLimit: 10G
diff --git a/tests/integration/kubernetes/tests_common.sh b/tests/integration/kubernetes/tests_common.sh
index f76566c07a..1f88877fd7 100644
--- a/tests/integration/kubernetes/tests_common.sh
+++ b/tests/integration/kubernetes/tests_common.sh
@@ -395,6 +395,30 @@ add_requests_to_policy_settings() {
 	done
 }
 
+# Change genpolicy settings to use the requested emptyDir storage type.
+# Appends a "replace" op to 99-test-overrides.json.
+set_genpolicy_emptydir_type() {
+	declare -r settings_dir="$1"
+	declare -r emptydir_type="$2"
+
+	auto_generate_policy_enabled || return 0
+
+	case "${emptydir_type}" in
+		shared-fs|block-encrypted|block-plain) ;;
+		*) die "Unsupported genpolicy emptydir_type ${emptydir_type}" ;;
+	esac
+
+	local drop_in_dir="${settings_dir}/genpolicy-settings.d"
+	mkdir -p "${drop_in_dir}"
+	local overrides_file="${drop_in_dir}/99-test-overrides.json"
+	[[ -f "${overrides_file}" ]] || echo '[]' > "${overrides_file}"
+
+	info "Setting genpolicy emptydir_type to ${emptydir_type} in ${overrides_file}"
+	jq --arg emptydir_type "${emptydir_type}" \
+		'. + [{"op":"replace","path":"/cluster_config/emptydir_type","value":$emptydir_type}]' \
+		"${overrides_file}" > "${overrides_file}.tmp" && mv "${overrides_file}.tmp" "${overrides_file}"
+}
+
 # Change Rego rules to allow one or more ttrpc requests from the Host to the Guest.
 allow_requests() {
 	declare -r settings_dir="$1"