CVE-2026-115883: CWE-20: Improper Input Validation in langchain-ai langgraph

LangGraph checkpoint savers for Postgres and SQLite flatten namespace labels by joining with dots, but only validate these labels on put and aput operations. Other operations such as get, search, delete, list_namespaces, and their asynchronous or batch equivalents do not validate dotted namespace labels. This allows an attacker to craft unvalidated dotted labels that can resolve to another tenant's namespace, leading to unauthorized cross-namespace reads, overwrites, or deletions. The InMemoryStore is unaffected because it compares namespace tuples directly. The vulnerability is fixed in langgraph-checkpoint-postgres version 3.1.3 and langgraph-checkpoint-sqlite version 3.1.2.

Exploitation of this vulnerability can result in unauthorized access to or modification of another tenant's stored namespace data within LangGraph workflows and agents using Postgres or SQLite checkpoint savers. This undermines data isolation between tenants, potentially leading to data leakage or corruption. The vulnerability does not affect InMemoryStore implementations.

A fix is available and should be applied by upgrading to langgraph-checkpoint-postgres version 3.1.3 or later and langgraph-checkpoint-sqlite version 3.1.2 or later. Users should update affected components to these versions to remediate the vulnerability.