diff --git a/rdagent/scenarios/rl/autorl_bench/benchmarks/webshop/requirements.txt b/rdagent/scenarios/rl/autorl_bench/benchmarks/webshop/requirements.txt
index 1acee7d1..7e3e4829 100644
--- a/rdagent/scenarios/rl/autorl_bench/benchmarks/webshop/requirements.txt
+++ b/rdagent/scenarios/rl/autorl_bench/benchmarks/webshop/requirements.txt
@@ -24,10 +24,10 @@ rank_bm25==0.2.2
 thefuzz==0.19.0
 spacy==3.7.2
 
-# Note: Flask/Werkzeug pinned to 2.x (Flask 3.x incompatible with WebShop)
+# Note: Flask/Werkzeug updated to 3.x for security fixes
 # Security Notes:
-# - CVE-2024-34069: Werkzeug debugger RCE (dev mode only)
-# - CVE-2024-49767: Resource exhaustion via multipart/form-data
-# Mitigation: Benchmark runs locally with max_content_length limits; never use debug=True in production
-flask==2.2.5
-Werkzeug==2.3.8  # Latest 2.x with security patches (CVE-2024-34069, CVE-2024-49767 mitigated)
\ No newline at end of file
+# - CVE-2025-66221: Windows device names in safe_join() (fixed in 3.1.4)
+# - CVE-2024-34069: Werkzeug debugger RCE (fixed)
+# - CVE-2024-49767: Resource exhaustion via multipart/form-data (fixed)
+flask>=3.0.0
+Werkzeug>=3.1.6  # Security fix: CVE-2025-66221 (Windows device names)
\ No newline at end of file