diff --git a/hscontrol/policy/v2/policy.go b/hscontrol/policy/v2/policy.go
index 6aeda271..c5d87722 100644
--- a/hscontrol/policy/v2/policy.go
+++ b/hscontrol/policy/v2/policy.go
@@ -498,8 +498,7 @@ func (pm *PolicyManager) SetNodes(nodes views.Slice[types.NodeView]) (bool, erro
 	pm.mu.Lock()
 	defer pm.mu.Unlock()
 
-	oldNodeCount := pm.nodes.Len()
-	newNodeCount := nodes.Len()
+	policyChanged := pm.nodesHavePolicyAffectingChanges(nodes)
 
 	// Invalidate cache entries for nodes that changed.
 	// For autogroup:self: invalidate all nodes belonging to affected users (peer changes).
@@ -508,19 +507,17 @@ func (pm *PolicyManager) SetNodes(nodes views.Slice[types.NodeView]) (bool, erro
 
 	pm.nodes = nodes
 
-	nodesChanged := oldNodeCount != newNodeCount
-
-	// When nodes are added/removed, we must recompile filters because:
+	// When policy-affecting node properties change, we must recompile filters because:
 	// 1. User/group aliases (like "user1@") resolve to node IPs
-	// 2. Filter compilation needs nodes to generate rules
-	// 3. Without nodes, filters compile to empty (0 rules)
+	// 2. Tag aliases (like "tag:server") match nodes based on their tags
+	// 3. Filter compilation needs nodes to generate rules
 	//
 	// For autogroup:self: return true when nodes change even if the global filter
 	// hash didn't change. The global filter is empty for autogroup:self (each node
 	// has its own filter), so the hash never changes. But peer relationships DO
 	// change when nodes are added/removed, so we must signal this to trigger updates.
 	// For global policies: the filter must be recompiled to include the new nodes.
-	if nodesChanged {
+	if policyChanged {
 		// Recompile filter with the new node list
 		needsUpdate, err := pm.updateLocked()
 		if err != nil {
@@ -541,6 +538,30 @@ func (pm *PolicyManager) SetNodes(nodes views.Slice[types.NodeView]) (bool, erro
 	return false, nil
 }
 
+func (pm *PolicyManager) nodesHavePolicyAffectingChanges(newNodes views.Slice[types.NodeView]) bool {
+	if pm.nodes.Len() != newNodes.Len() {
+		return true
+	}
+
+	oldNodes := make(map[types.NodeID]types.NodeView, pm.nodes.Len())
+	for _, node := range pm.nodes.All() {
+		oldNodes[node.ID()] = node
+	}
+
+	for _, newNode := range newNodes.All() {
+		oldNode, exists := oldNodes[newNode.ID()]
+		if !exists {
+			return true
+		}
+
+		if newNode.HasPolicyChange(oldNode) {
+			return true
+		}
+	}
+
+	return false
+}
+
 // NodeCanHaveTag checks if a node can have the specified tag during client-initiated
 // registration or reauth flows (e.g., tailscale up --advertise-tags).
 //
diff --git a/hscontrol/policy/v2/policy_test.go b/hscontrol/policy/v2/policy_test.go
index b348c0ab..f3f514a6 100644
--- a/hscontrol/policy/v2/policy_test.go
+++ b/hscontrol/policy/v2/policy_test.go
@@ -606,3 +606,126 @@ func TestAutogroupSelfPolicyUpdateTriggersMapResponse(t *testing.T) {
 	require.NoError(t, err)
 	require.False(t, policyChanged2, "SetPolicy should return false when policy content hasn't changed")
 }
+
+// TestTagPropagationToPeerMap tests that when a node's tags change,
+// the peer map is correctly updated. This is a regression test for
+// https://github.com/juanfont/headscale/issues/2389
+func TestTagPropagationToPeerMap(t *testing.T) {
+	users := types.Users{
+		{Model: gorm.Model{ID: 1}, Name: "user1", Email: "user1@headscale.net"},
+		{Model: gorm.Model{ID: 2}, Name: "user2", Email: "user2@headscale.net"},
+	}
+
+	// Policy: user2 can access tag:web nodes
+	policy := `{
+		"tagOwners": {
+			"tag:web": ["user1@headscale.net"],
+			"tag:internal": ["user1@headscale.net"]
+		},
+		"acls": [
+			{
+				"action": "accept",
+				"src": ["user2@headscale.net"],
+				"dst": ["user2@headscale.net:*"]
+			},
+			{
+				"action": "accept",
+				"src": ["user2@headscale.net"],
+				"dst": ["tag:web:*"]
+			},
+			{
+				"action": "accept",
+				"src": ["tag:web"],
+				"dst": ["user2@headscale.net:*"]
+			}
+		]
+	}`
+
+	// user1's node starts with tag:web and tag:internal
+	user1Node := &types.Node{
+		ID:       1,
+		Hostname: "user1-node",
+		IPv4:     ap("100.64.0.1"),
+		IPv6:     ap("fd7a:115c:a1e0::1"),
+		User:     ptr.To(users[0]),
+		UserID:   ptr.To(users[0].ID),
+		Tags:     []string{"tag:web", "tag:internal"},
+	}
+
+	// user2's node (no tags)
+	user2Node := &types.Node{
+		ID:       2,
+		Hostname: "user2-node",
+		IPv4:     ap("100.64.0.2"),
+		IPv6:     ap("fd7a:115c:a1e0::2"),
+		User:     ptr.To(users[1]),
+		UserID:   ptr.To(users[1].ID),
+	}
+
+	initialNodes := types.Nodes{user1Node, user2Node}
+
+	pm, err := NewPolicyManager([]byte(policy), users, initialNodes.ViewSlice())
+	require.NoError(t, err)
+
+	// Initial state: user2 should see user1 as a peer (user1 has tag:web)
+	initialPeerMap := pm.BuildPeerMap(initialNodes.ViewSlice())
+
+	// Check user2's peers - should include user1
+	user2Peers := initialPeerMap[user2Node.ID]
+	require.Len(t, user2Peers, 1, "user2 should have 1 peer initially (user1 with tag:web)")
+	require.Equal(t, user1Node.ID, user2Peers[0].ID(), "user2's peer should be user1")
+
+	// Check user1's peers - should include user2 (bidirectional ACL)
+	user1Peers := initialPeerMap[user1Node.ID]
+	require.Len(t, user1Peers, 1, "user1 should have 1 peer initially (user2)")
+	require.Equal(t, user2Node.ID, user1Peers[0].ID(), "user1's peer should be user2")
+
+	// Now change user1's tags: remove tag:web, keep only tag:internal
+	user1NodeUpdated := &types.Node{
+		ID:       1,
+		Hostname: "user1-node",
+		IPv4:     ap("100.64.0.1"),
+		IPv6:     ap("fd7a:115c:a1e0::1"),
+		User:     ptr.To(users[0]),
+		UserID:   ptr.To(users[0].ID),
+		Tags:     []string{"tag:internal"}, // tag:web removed!
+	}
+
+	updatedNodes := types.Nodes{user1NodeUpdated, user2Node}
+
+	// SetNodes should detect the tag change
+	changed, err := pm.SetNodes(updatedNodes.ViewSlice())
+	require.NoError(t, err)
+	require.True(t, changed, "SetNodes should return true when tags change")
+
+	// After tag change: user2 should NOT see user1 as a peer anymore
+	// (no ACL allows user2 to access tag:internal)
+	updatedPeerMap := pm.BuildPeerMap(updatedNodes.ViewSlice())
+
+	// Check user2's peers - should be empty now
+	user2PeersAfter := updatedPeerMap[user2Node.ID]
+	require.Empty(t, user2PeersAfter, "user2 should have no peers after tag:web is removed from user1")
+
+	// Check user1's peers - should also be empty
+	user1PeersAfter := updatedPeerMap[user1Node.ID]
+	require.Empty(t, user1PeersAfter, "user1 should have no peers after tag:web is removed")
+
+	// Also verify MatchersForNode returns non-empty matchers and ReduceNodes filters correctly
+	// This simulates what buildTailPeers does in the mapper
+	matchersForUser2, err := pm.MatchersForNode(user2Node.View())
+	require.NoError(t, err)
+	require.NotEmpty(t, matchersForUser2, "MatchersForNode should return non-empty matchers (at least self-access rule)")
+
+	// Test ReduceNodes logic with the updated nodes and matchers
+	// This is what buildTailPeers does - it takes peers from ListPeers (which might include user1)
+	// and filters them using ReduceNodes with the updated matchers
+	// Inline the ReduceNodes logic to avoid import cycle
+	user2View := user2Node.View()
+	user1UpdatedView := user1NodeUpdated.View()
+
+	// Check if user2 can access user1 OR user1 can access user2
+	canAccess := user2View.CanAccess(matchersForUser2, user1UpdatedView) ||
+		user1UpdatedView.CanAccess(matchersForUser2, user2View)
+
+	require.False(t, canAccess, "user2 should NOT be able to access user1 after tag:web is removed (ReduceNodes should filter out)")
+}
diff --git a/hscontrol/state/node_store_test.go b/hscontrol/state/node_store_test.go
index b9b8fcf2..3d6184ba 100644
--- a/hscontrol/state/node_store_test.go
+++ b/hscontrol/state/node_store_test.go
@@ -1152,3 +1152,92 @@ func TestNodeStoreAllocationStats(t *testing.T) {
 	allocs := res.AllocsPerOp()
 	t.Logf("NodeStore allocations per op: %.2f", float64(allocs))
 }
+
+// TestRebuildPeerMapsWithChangedPeersFunc tests that RebuildPeerMaps correctly
+// rebuilds the peer map when the peersFunc behavior changes.
+// This simulates what happens when SetNodeTags changes node tags and the
+// PolicyManager's matchers are updated, requiring the peer map to be rebuilt.
+func TestRebuildPeerMapsWithChangedPeersFunc(t *testing.T) {
+	// Create a peersFunc that can be controlled via a channel
+	// Initially it returns all nodes as peers, then we change it to return no peers
+	allowPeers := true
+
+	// This simulates how PolicyManager.BuildPeerMap works - it reads state
+	// that can change between calls
+	dynamicPeersFunc := func(nodes []types.NodeView) map[types.NodeID][]types.NodeView {
+		ret := make(map[types.NodeID][]types.NodeView, len(nodes))
+		if allowPeers {
+			// Allow all peers
+			for _, node := range nodes {
+				var peers []types.NodeView
+
+				for _, n := range nodes {
+					if n.ID() != node.ID() {
+						peers = append(peers, n)
+					}
+				}
+
+				ret[node.ID()] = peers
+			}
+		} else {
+			// Allow no peers
+			for _, node := range nodes {
+				ret[node.ID()] = []types.NodeView{}
+			}
+		}
+
+		return ret
+	}
+
+	// Create nodes
+	node1 := createTestNode(1, 1, "user1", "node1")
+	node2 := createTestNode(2, 2, "user2", "node2")
+	initialNodes := types.Nodes{&node1, &node2}
+
+	// Create store with dynamic peersFunc
+	store := NewNodeStore(initialNodes, dynamicPeersFunc, TestBatchSize, TestBatchTimeout)
+
+	store.Start()
+	defer store.Stop()
+
+	// Initially, nodes should see each other as peers
+	snapshot := store.data.Load()
+	require.Len(t, snapshot.peersByNode[1], 1, "node1 should have 1 peer initially")
+	require.Len(t, snapshot.peersByNode[2], 1, "node2 should have 1 peer initially")
+	require.Equal(t, types.NodeID(2), snapshot.peersByNode[1][0].ID())
+	require.Equal(t, types.NodeID(1), snapshot.peersByNode[2][0].ID())
+
+	// Now "change the policy" by disabling peers
+	allowPeers = false
+
+	// Call RebuildPeerMaps to rebuild with the new behavior
+	store.RebuildPeerMaps()
+
+	// After rebuild, nodes should have no peers
+	snapshot = store.data.Load()
+	assert.Empty(t, snapshot.peersByNode[1], "node1 should have no peers after rebuild")
+	assert.Empty(t, snapshot.peersByNode[2], "node2 should have no peers after rebuild")
+
+	// Verify that ListPeers returns the correct result
+	peers1 := store.ListPeers(1)
+	peers2 := store.ListPeers(2)
+
+	assert.Equal(t, 0, peers1.Len(), "ListPeers for node1 should return empty")
+	assert.Equal(t, 0, peers2.Len(), "ListPeers for node2 should return empty")
+
+	// Now re-enable peers and rebuild again
+	allowPeers = true
+
+	store.RebuildPeerMaps()
+
+	// Nodes should see each other again
+	snapshot = store.data.Load()
+	require.Len(t, snapshot.peersByNode[1], 1, "node1 should have 1 peer after re-enabling")
+	require.Len(t, snapshot.peersByNode[2], 1, "node2 should have 1 peer after re-enabling")
+
+	peers1 = store.ListPeers(1)
+	peers2 = store.ListPeers(2)
+
+	assert.Equal(t, 1, peers1.Len(), "ListPeers for node1 should return 1")
+	assert.Equal(t, 1, peers2.Len(), "ListPeers for node2 should return 1")
+}
diff --git a/hscontrol/types/node.go b/hscontrol/types/node.go
index a92382c4..e115df51 100644
--- a/hscontrol/types/node.go
+++ b/hscontrol/types/node.go
@@ -973,6 +973,23 @@ func (nv NodeView) HasNetworkChanges(other NodeView) bool {
 	return false
 }
 
+// HasPolicyChange reports whether the node has changes that affect policy evaluation.
+func (nv NodeView) HasPolicyChange(other NodeView) bool {
+	if nv.UserID() != other.UserID() {
+		return true
+	}
+
+	if !views.SliceEqual(nv.Tags(), other.Tags()) {
+		return true
+	}
+
+	if !slices.Equal(nv.IPs(), other.IPs()) {
+		return true
+	}
+
+	return false
+}
+
 // TailNodes converts a slice of NodeViews into Tailscale tailcfg.Nodes.
 func TailNodes(
 	nodes views.Slice[NodeView],