LangGraph Flaw Chain Exposes Self-Hosted AI Agents to Remote Code Execution

Three patched security vulnerabilities in LangGraph, an open-source AI agentic framework by LangChain, have been publicly disclosed, with a critical flaw chain enabling remote code execution.

• •CVE-2025-67644 (CVSS 7.3) exploits SQL injection in SQLite checkpoint implementation to manipulate SQL queries through metadata filter keys
• •CVE-2026-28277 (CVSS 6.8) leverages unsafe msgpack deserialization to execute code when checkpoints containing attacker-controlled data are loaded
• •CVE-2026-27022 (CVSS 6.5) abuses RediSearch Query Injection to bypass access controls in Redis checkpoint implementation
• •The vulnerability chain uses SQL injection to inject fake checkpoints with malicious serialized payloads, then triggers unsafe deserialization for arbitrary code execution
• •