LangChain, LangGraph Flaws Expose Files, Secrets, Databases in Widely Used AI Frameworks

**Executive summary:** Cybersecurity researchers disclosed three serious vulnerabilities in LangChain and LangGraph — a path traversal (CVE-2026-34070), unsafe deserialization leaking API keys and secrets (CVE-2025-68664), and an SQL injection in LangGraph's SQLite checkpoint (CVE-2025-67644) — that could allow attackers to read files, siphon secrets, and access conversation history; patches for the affected versions have been released and users are urged to update promptly.