Cybersecurity researchers have identified three critical vulnerabilities in the LangGraph AI framework that could allow attackers to achieve remote code execution through a chained exploit of database flaws.

Key Points

• Researchers discovered three vulnerabilities, including CVE-2025-67644, CVE-2026-28277, and CVE-2026-27022, affecting LangGraph's SQLite and Redis checkpoint implementations.
• A critical attack chain allows unauthorized users to combine SQL injection and unsafe msgpack deserialization to execute arbitrary code on self-hosted servers.
• The vulnerabilities specifically impact versions of langgraph-checkpoint-sqlite before 3.0.1, langgraph before 1.0.10, and @langchain/langgraph-checkpoint-redis before 1.0.1.
• LangChain confirmed that its managed LangSmith deployment platform remains unaffected by these security flaws.
• Security experts recommend that users update to the latest patched versions and implement strict network segmentation and authentication for all self-hosted AI agent deployments.

Why it Matters

These vulnerabilities highlight the significant security risks associated with integrating traditional database flaws into complex AI agent frameworks that possess elevated system privileges. Organizations must prioritize patching and secure configuration to prevent attackers from exploiting these frameworks to gain unauthorized access to sensitive runtime secrets and internal systems.