diff --git a/CHANGELOG.md b/CHANGELOG.md
index 21835a5c3..956765c6a 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -5,6 +5,9 @@ The format is based on [keep a changelog](http://keepachangelog.com) and this pr
 
 ## [Unreleased]
 
+### Fixed
+- Ignore negative custom runtime counter deltas to avoid panics during metrics collection.
+
 ## [3.38.0] - 2026-03-20
 ### Added
 - Add runtime Satori client feature to delete identities.
diff --git a/server/metrics.go b/server/metrics.go
index 9fbfee516..d2c0bbe1f 100644
--- a/server/metrics.go
+++ b/server/metrics.go
@@ -92,10 +92,11 @@ type LocalMetrics struct {
 	currentRecvBytes *atomic.Int64
 	currentSentBytes *atomic.Int64
 
-	PrometheusScope       tally.Scope
-	prometheusCustomScope tally.Scope
-	prometheusCloser      io.Closer
-	prometheusHTTPServer  *http.Server
+	PrometheusScope        tally.Scope
+	prometheusCustomScope  tally.Scope
+	metricsCollectionScope tally.Scope
+	prometheusCloser       io.Closer
+	prometheusHTTPServer   *http.Server
 }
 
 func NewLocalMetrics(logger, startupLogger *zap.Logger, db *sql.DB, config Config) *LocalMetrics {
@@ -162,8 +163,10 @@ func NewLocalMetrics(logger, startupLogger *zap.Logger, db *sql.DB, config Confi
 		SanitizeOptions: &prometheus.DefaultSanitizerOpts,
 	}, time.Duration(config.GetMetrics().ReportingFreqSec)*time.Second)
 	m.prometheusCustomScope = m.PrometheusScope.SubScope(config.GetMetrics().CustomPrefix)
+	m.metricsCollectionScope = m.PrometheusScope
 	if config.GetMetrics().CustomScopeLimit > 0 {
 		m.prometheusCustomScope = newMetricsLimitedScope(m.prometheusCustomScope, int64(config.GetMetrics().CustomScopeLimit))
+		m.metricsCollectionScope = newMetricsLimitedScope(m.metricsCollectionScope, int64(config.GetMetrics().CustomScopeLimit))
 	}
 
 	// Check if exposing Prometheus metrics directly is enabled.
@@ -523,6 +526,11 @@ func (m *LocalMetrics) StorageWriteRejectCount(tags map[string]string, delta int
 
 // CustomCounter adds the given delta to a counter with the specified name and tags.
 func (m *LocalMetrics) CustomCounter(name string, tags map[string]string, delta int64) {
+	if delta < 0 {
+		m.metricsCollectionScope.Tagged(map[string]string{"name": name, "err": "negative_increment"}).Counter("metrics_collection_errors_count").Inc(1)
+		return
+	}
+
 	scope := m.prometheusCustomScope
 	if len(tags) != 0 {
 		scope = scope.Tagged(tags)
diff --git a/server/metrics_test.go b/server/metrics_test.go
new file mode 100644
index 000000000..215c77840
--- /dev/null
+++ b/server/metrics_test.go
@@ -0,0 +1,41 @@
+// Copyright 2026 The Nakama Authors
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+package server
+
+import (
+	"testing"
+	"time"
+
+	"go.uber.org/zap"
+)
+
+func TestMetricsCounterAddNegativeDoesNotPanic(t *testing.T) {
+	logger := zap.NewNop()
+	cfg := NewConfig(logger)
+	cfg.Metrics.ReportingFreqSec = 1
+	reportingInterval := time.Duration(cfg.Metrics.ReportingFreqSec) * time.Second
+	flushWait := reportingInterval + 200*time.Millisecond
+
+	metrics := NewLocalMetrics(logger, logger, nil, cfg)
+	defer metrics.Stop(logger)
+
+	module := &RuntimeGoNakamaModule{metrics: metrics}
+	module.MetricsCounterAdd("panic_counter", nil, 1)
+
+	time.Sleep(flushWait)
+	module.MetricsCounterAdd("panic_counter", nil, -1)
+
+	time.Sleep(flushWait)
+}