Skip to the content
AI and Cybersecurity Learning and Practice Center Test automation Open the Learning Center

Is my AI tool safe?Agent frameworks and buildersLangChain

Are LangChain and LangGraph safe?

No tool is safe or unsafe on its own: it depends on what it can reach and how it is set up. The record for LangChain and LangGraph: 0 incidents in this center's library and 6 CVEs in its advisory feed (2 critical, 4 high), the newest published Mar 27, 2026. Below: what went wrong, what was fixed, and how to use it safely.

What to watch for

The most-installed agent framework. No headline incident in the library, but a steady run of CVEs in serialization (LangGrinch), legacy prompt loading and checkpoint storage, each reachable from content a chain processes.

How to use it safely

Pin the whole family (core, community, checkpoint packages), re-check on every advisory, and never deserialize checkpoints or templates from untrusted sources.

What it can reach

  • Reads attacker-controlled content
  • Loads model files that can execute code

Known CVEs

  • CVE-2026-34070 High LangChain path traversal in legacy load_prompt functions
    Mar 27, 2026 · fixed in 1.2.22
  • CVE-2025-68664 Critical LangChain serialization injection exposes secrets (LangGrinch)
    Dec 23, 2025 · no fixed release in the feed
  • CVE-2025-68665 High LangChain.js serialization injection exposes secrets
    Dec 23, 2025 · no fixed release in the feed
  • CVE-2025-67644 High LangGraph SQLite checkpointer SQL injection
    Dec 10, 2025 · fixed in 3.0.1
  • CVE-2025-64439 High LangGraph checkpoint deserialization remote code execution
    Nov 5, 2025 · fixed in 3.0.0
  • CVE-2023-36258 Critical LangChain arbitrary code execution through PALChain
    Jul 3, 2023 · fixed in 0.0.247

How many people use it

  • 169M PyPI downloads of langchain in a month (to Oct 1, 2026) · PyPI Stats
  • 145M PyPI downloads of langchain-core in a month · PyPI Stats
  • 147k GitHub stars · GitHub

A high CVE count can mean heavy scrutiny as much as weak engineering, so read the incidents and advisories, not only the number. The counts are what this center has recorded, not every flaw ever found; usage figures are the vendors' and registries' own published numbers.

Sources

Compare it with the others on the Technology watch list