CVE-2025-64439: RCE Flaw Detected in LangGraph: Agent Orchestration Framework at Risk

CVE-2025-64439 is an RCE flaw detected in LangGraph, a low-level agent orchestration framework used for building stateful AI agents, with a CVSS of 7.4. The vulnerability resides in the JsonPlusSerializer component and could allow an attacker to execute arbitrary Python code on affected systems, according to the advisory GHSA-wwqv-p2pp-99h5.

It stems from a dangerous fallback in the LangGraph checkpoint serializer: prior to version 3.0, if serialization failed due to illegal Unicode surrogate values, the system could fall back to a json mode where the deserializer supported a constructor-style format to rebuild objects, enabling code execution on load.

The issue could permit full remote code execution for applications that persist untrusted data into checkpoints, affecting all users of langgraph-checkpoint versions earlier than 3.0; risk is higher where untrusted data is accepted and the default serializer may fall back to json mode. LangGraph has issued a fix in version 3.0, and upgrading is strongly advised for users deploying via langgraph-api.