GHSA-MHR3-J7M5-C7C9

Vulnerability from github – Published: 2026-02-25 22:59 – Updated: 2026-02-25 22:59
VLAI
Summary
LangGraph: BaseCache Deserialization of Untrusted Data may lead to Remote Code Execution
Details

Context

A Remote Code Execution vulnerability exists in LangGraph's caching layer when applications enable cache backends that inherit from BaseCache and opt nodes into caching via CachePolicy. Prior to langgraph-checkpoint 4.0.0, BaseCache defaults to JsonPlusSerializer(pickle_fallback=True). When msgpack serialization fails, cached values can be deserialized via pickle.loads(...).

Who is affected?

Caching is not enabled by default. Applications are affected only when:

  • The application explicitly enables a cache backend (for example by passing cache=... to StateGraph.compile(...) or otherwise configuring a BaseCache implementation)
  • One or more nodes opt into caching via CachePolicy
  • The attacker can write to the cache backend (for example a network-accessible Redis instance with weak/no auth, shared cache infrastructure reachable by other tenants/services, or a writable SQLite cache file)

Example (enabling a cache backend and opting a node into caching):

from langgraph.cache.memory import InMemoryCache
from langgraph.graph import StateGraph
from langgraph.types import CachePolicy


def my_node(state: dict) -> dict:
    return {"value": state.get("value", 0) + 1}


builder = StateGraph(dict)
builder.add_node("my_node", my_node, cache_policy=CachePolicy(ttl=120))
builder.set_entry_point("my_node")

graph = builder.compile(cache=InMemoryCache())

result = graph.invoke({"value": 1})

With pickle_fallback=True, when msgpack serialization fails, JsonPlusSerializer can fall back to storing values as a ("pickle", <bytes>) tuple and later deserialize them via pickle.loads(...). If an attacker can place a malicious pickle payload into the cache backend such that the LangGraph process reads and deserializes it, this can lead to arbitrary code execution.

Exploitation requires attacker write access to the cache backend. The serializer is not exposed as a network-facing API.

This is fixed in langgraph-checkpoint>=4.0.0 by disabling pickle fallback by default (pickle_fallback=False).

Impact

Arbitrary code execution in the LangGraph process when attacker-controlled cache entries are deserialized.

Root Cause

  • BaseCache default serializer configuration inherited by cache implementations (InMemoryCache, RedisCache, SqliteCache):
  • libs/checkpoint/langgraph/cache/base/__init__.py (pre-fix default: JsonPlusSerializer(pickle_fallback=True))

  • JsonPlusSerializer deserialization sink:

  • libs/checkpoint/langgraph/checkpoint/serde/jsonplus.py
  • loads_typed(...) calls pickle.loads(data_) when type_ == "pickle" and pickle fallback is enabled

Attack preconditions

An attacker must be able to write attacker-controlled bytes into the cache backend such that the LangGraph process later reads and deserializes them.

This typically requires write access to a networked cache (for example a network-accessible Redis instance with weak/no auth or shared cache infrastructure reachable by other tenants/services) or write access to local cache storage (for example a writable SQLite cache file via permissive file permissions or a shared writable volume).

Because exploitation requires write access to the cache storage layer, this is a post-compromise / post-access escalation vector.

Remediation

  • Upgrade to langgraph-checkpoint>=4.0.0.

Resources

  • ZDI-CAN-28385
  • Patch: https://github.com/langchain-ai/langgraph/pull/6677
  • Patch diff: https://patch-diff.githubusercontent.com/raw/langchain-ai/langgraph/pull/6677.patch
  • Credit: Peter Girnus (@gothburz), Demeng Chen, and Brandon Niemczyk (Trend Micro Zero Day Initiative)
Show details on source website

{
  "affected": [
    {
      "package": {
        "ecosystem": "PyPI",
        "name": "langgraph-checkpoint"
      },
      "ranges": [
        {
          "events": [
            {
              "introduced": "0"
            },
            {
              "fixed": "4.0.0"
            }
          ],
          "type": "ECOSYSTEM"
        }
      ]
    }
  ],
  "aliases": [
    "CVE-2026-27794"
  ],
  "database_specific": {
    "cwe_ids": [
      "CWE-502"
    ],
    "github_reviewed": true,
    "github_reviewed_at": "2026-02-25T22:59:12Z",
    "nvd_published_at": "2026-02-25T18:23:40Z",
    "severity": "MODERATE"
  },
  "details": "## Context\n\nA Remote Code Execution vulnerability exists in LangGraph\u0027s caching layer when applications enable cache backends that inherit from `BaseCache` and opt nodes into caching via `CachePolicy`. Prior to `langgraph-checkpoint` 4.0.0, `BaseCache` defaults to `JsonPlusSerializer(pickle_fallback=True)`. When msgpack serialization fails, cached values can be deserialized via `pickle.loads(...)`.\n\n### Who is affected?\n\nCaching is not enabled by default. Applications are affected only when:\n\n- The application explicitly enables a cache backend (for example by passing `cache=...` to `StateGraph.compile(...)` or otherwise configuring a `BaseCache` implementation)\n- One or more nodes opt into caching via `CachePolicy`\n- The attacker can write to the cache backend (for example a network-accessible Redis instance with weak/no auth, shared cache infrastructure reachable by other tenants/services, or a writable SQLite cache file)\n\nExample (enabling a cache backend and opting a node into caching):\n\n```py\nfrom langgraph.cache.memory import InMemoryCache\nfrom langgraph.graph import StateGraph\nfrom langgraph.types import CachePolicy\n\n\ndef my_node(state: dict) -\u003e dict:\n    return {\"value\": state.get(\"value\", 0) + 1}\n\n\nbuilder = StateGraph(dict)\nbuilder.add_node(\"my_node\", my_node, cache_policy=CachePolicy(ttl=120))\nbuilder.set_entry_point(\"my_node\")\n\ngraph = builder.compile(cache=InMemoryCache())\n\nresult = graph.invoke({\"value\": 1})\n```\n\nWith `pickle_fallback=True`, when msgpack serialization fails, `JsonPlusSerializer` can fall back to storing values as a `(\"pickle\", \u003cbytes\u003e)` tuple and later deserialize them via `pickle.loads(...)`. If an attacker can place a malicious pickle payload into the cache backend such that the LangGraph process reads and deserializes it, this can lead to arbitrary code execution.\n\nExploitation requires attacker write access to the cache backend. The serializer is not exposed as a network-facing API.\n\nThis is fixed in `langgraph-checkpoint\u003e=4.0.0` by disabling pickle fallback by default (`pickle_fallback=False`).\n\n## Impact\n\nArbitrary code execution in the LangGraph process when attacker-controlled cache entries are deserialized.\n\n## Root Cause\n\n- `BaseCache` default serializer configuration inherited by cache implementations (`InMemoryCache`, `RedisCache`, `SqliteCache`):\n  - `libs/checkpoint/langgraph/cache/base/__init__.py` (pre-fix default: `JsonPlusSerializer(pickle_fallback=True)`)\n\n- `JsonPlusSerializer` deserialization sink:\n  - `libs/checkpoint/langgraph/checkpoint/serde/jsonplus.py`\n  - `loads_typed(...)` calls `pickle.loads(data_)` when `type_ == \"pickle\"` and pickle fallback is enabled\n\n## Attack preconditions\n\nAn attacker must be able to write attacker-controlled bytes into the cache backend such that the LangGraph process later reads and deserializes them.\n\nThis typically requires write access to a networked cache (for example a network-accessible Redis instance with weak/no auth or shared cache infrastructure reachable by other tenants/services) or write access to local cache storage (for example a writable SQLite cache file via permissive file permissions or a shared writable volume).\n\nBecause exploitation requires write access to the cache storage layer, this is a post-compromise / post-access escalation vector.\n\n## Remediation\n\n- Upgrade to `langgraph-checkpoint\u003e=4.0.0`.\n\n## Resources\n\n- ZDI-CAN-28385\n- Patch: https://github.com/langchain-ai/langgraph/pull/6677\n- Patch diff: https://patch-diff.githubusercontent.com/raw/langchain-ai/langgraph/pull/6677.patch\n- Credit: Peter Girnus (@gothburz), Demeng Chen, and Brandon Niemczyk (Trend Micro Zero Day Initiative)",
  "id": "GHSA-mhr3-j7m5-c7c9",
  "modified": "2026-02-25T22:59:12Z",
  "published": "2026-02-25T22:59:12Z",
  "references": [
    {
      "type": "WEB",
      "url": "https://github.com/langchain-ai/langgraph/security/advisories/GHSA-mhr3-j7m5-c7c9"
    },
    {
      "type": "ADVISORY",
      "url": "https://nvd.nist.gov/vuln/detail/CVE-2026-27794"
    },
    {
      "type": "WEB",
      "url": "https://github.com/langchain-ai/langgraph/pull/6677"
    },
    {
      "type": "WEB",
      "url": "https://github.com/langchain-ai/langgraph/commit/f91d79d0c86932ded6e3b9f195d5a0bbd5aef99c"
    },
    {
      "type": "PACKAGE",
      "url": "https://github.com/langchain-ai/langgraph"
    },
    {
      "type": "WEB",
      "url": "https://github.com/langchain-ai/langgraph/releases/tag/checkpoint%3D%3D4.0.0"
    }
  ],
  "schema_version": "1.4.0",
  "severity": [
    {
      "score": "CVSS:3.1/AV:N/AC:H/PR:H/UI:N/S:U/C:H/I:H/A:H",
      "type": "CVSS_V3"
    }
  ],
  "summary": "LangGraph: BaseCache Deserialization of Untrusted Data may lead to Remote Code Execution "
}



Log in or create an account to share your comment.




Tags
Taxonomy of the tags.


Loading…

Loading…

Loading…

Drag nodes to arrange them. Double-click a related vulnerability to open it, or a source to visit its link. The Physics button on the left tunes the layout and the simulation; the legend at the bottom left hides or shows nodes by kind and by sighting type.

Forecast uses a logistic model when the trend is rising, or an exponential decay model when the trend is falling. Fitted via linearized least squares.

Sightings

Author Source Type Date Other

Nomenclature

  • Seen: The vulnerability was mentioned, discussed, or observed by the user.
  • Confirmed: The vulnerability has been validated from an analyst's perspective.
  • Published Proof of Concept: A public proof of concept is available for this vulnerability.
  • Exploited: The vulnerability was observed as exploited by the user who reported the sighting.
  • Patched: The vulnerability was observed as successfully patched by the user who reported the sighting.
  • Not exploited: The vulnerability was not observed as exploited by the user who reported the sighting.
  • Not confirmed: The user expressed doubt about the validity of the vulnerability.
  • Not patched: The vulnerability was not observed as successfully patched by the user who reported the sighting.
Vulnerability timeline

Sightings and recorded dates, grouped by day · UTC

JSON
How durations are calculated

Following the VTEM framework (Black Hat USA 2015), time to patch starts when the vendor or a coordinator was informed, total time to patch at discovery, and disclosure to patch at public disclosure; vendor response time runs from the report to its acknowledgement. Time to exploit and time unpatched start at the earliest discovery or report, falling back to public disclosure. Unpatched time ends at the first recorded patch, or the calculation time when no patch is recorded. Exploitation to patch shows the interval between those two events; a published proof of concept is a public exploit, not exploitation.

Each result shows its source dates. A negative interval means the end event came first; when date-only inputs leave the order open, the result says so. Date-only inputs are approximate. The earliest evidence wins: KEV report dates and sightings are labelled as such. Records describing another vulnerability are listed but never calculated from, and a PoC alone does not establish exploitation. Missing patch evidence does not prove that no patch exists. Filters below change the event list only.

Events included in the timeline

Loading timeline…


    Loading…

    Loading…

    Loading…

    Related by attack behaviour

    Vulnerabilities whose description is nearest to this one in the vector space of the CIRCL/vulnerability-attack-technique-biencoder model. This is a similarity search over the bi-encoder space (plain cosine), not a classification, and it has no measured accuracy.