LangChain and LangGraph, two popular open source frameworks used by over 60 million developers weekly, patched three high-severity vulnerabilities that expose files, API keys, and conversation histories. Security researchers said each vulnerability targets a different class of enterprise data, creating multiple attack vectors for threat actors.
The three flaws include CVE-2026-34070, a path traversal bug in LangChain that enables arbitrary file access without validation, CVE-2025-68664, a critical deserialization flaw allowing API key and environment secret exfiltration through prompt injection, and CVE-2025-67644, an SQL injection vulnerability in LangGraph’s SQLite checkpoint implementation enabling query manipulation.
Cyera security researcher Vladimir Tokarev explained that exploiting these flaws allows threat actors to read sensitive files including Docker configurations, exfiltrate secrets through prompt injection, and access conversation histories associated with sensitive workflows. LangChain serves as a foundational framework connecting AI models to data sources and tools for building chatbots and AI assistants, while LangGraph enables creation of AI agents following structured workflows.
The critical concern extends beyond the affected frameworks themselves. LangChain sits at the center of a massive dependency web stretching across the AI stack, with hundreds of libraries wrapping, extending, or depending on it. Researchers warn that vulnerabilities ripple outward through every downstream library, wrapper, and integration inheriting the vulnerable code paths.
Developers can fix CVE-2026-34070 by updating langchain-core to version 1.2.22, CVE-2025-68664 by updating to versions 0.3.81 and 1.2.5, and CVE-2025-67644 by updating langgraph-checkpoint-sqlite to version 3.0.1. Security experts urge auditing any code passing external or user-controlled configurations to load functions, disabling secrets_from_env defaults, and treating all LLM outputs as untrusted input since prompt injection can influence different fields.
