CVE-2026-118667: LangGraph: Unencoded identifiers in LangGraph SDK v3 streaming can redirect authenticated GET requests
The LangGraph Python SDK (`langgraph-sdk` 0.4.0 to 0.4.5) puts `thread_id` and `assistant_id` into streaming request paths without URL-encoding. If an app passes attacker-controlled text as one of them, the attacker can point an authenticated GET at another path on the same LangGraph server. Fixed in `langgraph-sdk` 0.4.6 and `langgraph` 1.2.14.
Published Oct 28, 2026 · Analysis Oct 29, 2026
What an attacker can do
In the SDK v3 streaming client, get_tree and _fetch_state build request paths from raw identifiers, at four call sites across the sync and async modules. They skip the SDK's own _quote_path_param helper. A value with relative segments such as ../ gets collapsed during HTTP path normalization before dispatch. The request still carries the client's authentication headers.
The effect is limited. The app must pass attacker-controlled text as a streaming thread_id or assistant_id. The attacker can then choose another readable path on the configured server, and the SDK returns that response to the application. Through thread_id they also control the query string, for example /store/items?namespace=...&key=.... Through assistant_id they control only the path. All the affected requests are GET, so there is no write primitive. LangGraph Platform is not exposed.
RemoteGraph v3 streaming in langgraph 1.2.3 and later is affected too when paired with a vulnerable SDK. It passes config["configurable"]["thread_id"] into threads.stream(...). A chat front end that lets users supply their own thread id is the typical risk. The attacker may read store items or other thread data that the app's credentials can read, but which they should not see.
Who is affected
langgraph-sdk 0.4.0 through 0.4.5. The v3 stream module does not exist in the 0.3.x line. langgraph 1.2.3 up to 1.2.13 is affected when it uses RemoteGraph v3 streaming with a vulnerable SDK. You are at risk only if user input reaches a streaming thread_id or assistant_id.
pip show langgraph-sdk langgraphlanggraph-sdk
- Ecosystem
- PyPI
- Affected
- >= 0.4.0, < 0.4.6
- Fixed in
- 0.4.6
- Check
pip show langgraph-sdk
langgraph
- Ecosystem
- PyPI
- Affected
- >= 1.2.3, < 1.2.14
- Fixed in
- 1.2.14
- Check
pip show langgraph
How to fix it
Upgrade langgraph-sdk to 0.4.6, the patched version:
pip install --upgrade "langgraph-sdk>=0.4.6"The fix runs assistant_id in get_tree and thread_id in _fetch_state through _quote_path_param, in the sync and async stream modules. If you use RemoteGraph, also upgrade langgraph to 1.2.14, which requires langgraph-sdk>=0.4.6:
pip install --upgrade "langgraph>=1.2.14"The advisory gives no other workaround. Validating ids on your side, for example accepting only UUIDs, also closes the hole.
How to tell if you were hit
Check your LangGraph server access logs for streaming-related GET requests whose path does not match the expected thread or assistant routes, especially paths to /store/items with namespace or key queries. In app logs, look for thread_id or assistant_id values that contain /, .., ? or %2e. Review which code paths pass user-supplied text into threads.stream(...) or into config["configurable"]["thread_id"].
Original advisory
From the GitHub Advisory Database (CC BY 4.0).
Summary
The LangGraph Python SDK v3 streaming client interpolates caller-supplied thread_id and assistant_id values into request paths without encoding them at four synchronous and asynchronous call sites. Relative path segments can therefore redirect an authenticated GET request to another path on the same configured server.
Details
The affected get_tree and _fetch_state implementations build request paths from raw identifiers rather than passing each identifier through the SDK's existing _quote_path_param helper. HTTP path normalization can collapse values containing relative segments before dispatch while retaining the client's authentication headers.
langgraph-sdk versions 0.4.0 through 0.4.5 are affected. The v3 stream module is not present in the 0.3.x release line.
RemoteGraph v3 streaming in langgraph 1.2.3 and later is also affected when used with a vulnerable langgraph-sdk version. It passes config["configurable"]["thread_id"] into threads.stream(...), and its dependency range permits affected SDK versions.
Impact
An application is affected only if it passes attacker-controlled text as a streaming thread_id or assistant_id. In that configuration, an attacker may select another readable path on the configured server and cause the SDK to return that response to the application. Through thread_id, the attacker can also control the query string, for example by targeting /store/items?namespace=...&key=...; assistant_id controls only the path. The affected requests are GET operations, so this issue does not provide a write primitive.
The LangGraph Platform is not exposed to this issue.
Remediation
Upgrade langgraph-sdk to version 0.4.6 or later. The fix encodes assistant_id in get_tree and thread_id in _fetch_state with _quote_path_param in both synchronous and asynchronous stream modules.
If you use RemoteGraph, also upgrade langgraph to version 1.2.14 or later, which requires langgraph-sdk>=0.4.6.
Credited to Evelynkaz.
Questions
- Is LangGraph Platform affected by CVE-2026-118667?
No. The advisory states the platform is not exposed. The flaw lives in the Python SDK client that applications run.
- Can an attacker write or delete data through this?
No. The affected calls are GET requests, so the advisory says there is no write primitive. The risk is reading another path on the same server.
- Does it matter if I use RemoteGraph?
Yes.
RemoteGraphv3 streaming inlanggraph1.2.3 and later passesthread_idto the SDK. Upgradelanggraphto 1.2.14 as well as the SDK.- Am I affected if thread ids are generated by my server?
The advisory says an app is affected only if it passes attacker-controlled text as a streaming
thread_idorassistant_id.
References
Read for this analysis:
- https://github.com/langchain-ai/langgraph/security/advisories/GHSA-3fx2-cqw9-xr3c
- https://github.com/langchain-ai/langgraph/pull/9213
- https://github.com/langchain-ai/langgraph/commit/d89c7e100ec6fc1512544d23b0a7c5169699ef3e
- https://github.com/langchain-ai/langgraph/releases/tag/1.2.14
- https://github.com/langchain-ai/langgraph/releases/tag/sdk%3D%3D0.4.6
How this page is made: the data panel comes from the GitHub Advisory Database, CISA’s KEV catalog, the NVD and FIRST EPSS, and updates twice a day. The analysis is written by an AI model from the sources listed above and checked by automated rules before it is published. Check versions against the vendor's advisory before you act. Methodology
More LangGraph vulnerabilities
All LangGraph entriesCVE-2026-48775
LangGraph Checkpoint: Unsafe JSON deserialization in checkpoint loading
medium6.8Jun 25, 2026CVE-2026-28277
LangGraph checkpoint loading has unsafe msgpack deserialization
medium6.8Mar 5, 2026CVE-2026-27794
LangGraph: BaseCache Deserialization of Untrusted Data may lead to Remote Code Execution
medium6.6Feb 25, 2026CVE-2025-64439
LangGraph Checkpoint affected by RCE in "json" mode of JsonPlusSerializer
high7.4Nov 5, 2025
