GHSA-8QPJ-27X8-PWPQ

Vulnerability from github – Published: 2026-10-06 13:38 – Updated: 2026-10-06 13:38
VLAI
Summary
Langflow: PythonREPLComponent executes unsandboxed Python code, enabling authenticated RCE and privilege escalation
Details

Summary

Langflow's built-in Python interpreter components — PythonREPLComponent (Python Interpreter) and the legacy PythonREPLToolComponent (Python REPL Tool) — executed arbitrary user- or model-supplied Python code inside flows without effective sandboxing. Because the code ran in-process with the privileges of the Langflow service, any authenticated user who could edit and run a flow could achieve remote code execution and, from there, escalate privileges to superuser (e.g. by opening a database session and flipping is_superuser) or compromise the host.

This issue is fixed as of 1.10.1, with additional hardening through 1.12.3. See Remediation below.

Affected

  • Package: langflow (PyPI), and the underlying lfx package that ships the component.
  • Vulnerable versions: < 1.10.1.
  • Patched: 1.10.1 (core fix). Upgrade to >= 1.12.3 for the complete hardening series.

Details

The root cause is code injection (CWE-94/CWE-95): the component passed raw input to LangChain's PythonREPL, which is explicitly not a security sandbox.

Two distinct weaknesses existed before 1.10.1:

  1. Unrestricted builtins (default deployments). get_globals() built the exec globals from the global_imports allow-list but never set __builtins__. CPython's exec() then auto-injected the full builtins module, leaving __import__, open, eval, exec and the whole import machinery reachable regardless of the allow-list — e.g. __import__("os").system(...) or __import__("subprocess").check_output([...]). This made the "only modules in Global Imports can be used" guarantee false, and it applied even with the default configuration (allow_custom_components=True).

  2. No server-policy gate (locked-down deployments). Even a deployment hardened with allow_custom_components=False could still run interpreter code, because the components did not consult that policy before executing.

Both let an authenticated user run the reported PoC, which opens a DB session and sets is_superuser = True on their account, or writes to the filesystem / runs OS commands with the service's privileges.

PoC (as reported)

  1. Authenticate as a normal user.
  2. Create a flow with the PythonREPLComponent.
  3. Execute Python that imports Langflow internals and elevates the account:
import asyncio
from sqlmodel import select
from langflow.services.database.models.user.model import User
from langflow.services.deps import session_scope

async def escalate():
    async with session_scope() as session:
        stmt = select(User).where(User.username == 'testuser')
        user = (await session.exec(stmt)).first()
        if user:
            user.is_superuser = True
            session.add(user)
            await session.commit()

asyncio.run(escalate())

Impact

Any authenticated user could: - Execute arbitrary Python / OS commands with the Langflow service's privileges (RCE). - Escalate their own account to superuser via direct database access. - Read/modify data and configuration, and potentially pivot to the underlying host.

Remediation

Upgrade to Langflow 1.10.1 or later (preferably >= 1.12.3). The interpreter components were hardened with layered, defense-in-depth controls, applied in run_python_repl() before any code is executed:

  • Restricted builtins — get_globals() injects a curated safe_builtins() mapping, removing __import__, eval, exec, compile, open, input, globals/locals/vars, getattr/setattr, etc. (#13397)
  • AST validation — validate_code_safety() rejects inline import/from ... import, dunder/escape-gadget attribute access (__class__, __subclasses__, __globals__, frame/traceback introspection) and format-string dunder traversal. (#13397)
  • Server-policy gate — ensure_code_execution_enabled() refuses to run when allow_custom_components=False or block_code_interpreter_components=True, and fails closed if the settings stack cannot be resolved. (#13700 — this advisory — and #14375)
  • Allow-listed module proxies — imported modules are exposed via a proxy that blocks reaching sys.modules["os"] through a module's transitive import graph. (#15198)
  • Optional hardware isolation — configure LANGFLOW_SANDBOX_BACKEND to run interpreter code in an isolated microVM instead of in-process. (#14400)

Upstream fix references: #13397, #13700 (carries this GHSA), #14375, #14400, #15198.

Hardening recommendations for operators

  • Keep Langflow updated (>= 1.12.3).
  • For locked-down deployments, set LANGFLOW_ALLOW_CUSTOM_COMPONENTS=false (or LANGFLOW_BLOCK_CODE_INTERPRETER_COMPONENTS=true) to disable the interpreter entirely.
  • For deployments that must run untrusted code, configure LANGFLOW_SANDBOX_BACKEND for microVM isolation.
  • Run the Langflow service as an unprivileged user with least-privilege database credentials.
Severity
9.9 (Critical)
CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:C/C:H/I:H/A:H
Show details on source website
To clipboard 

{
  "affected": [
    {
      "package": {
        "ecosystem": "PyPI",
        "name": "langflow"
      },
      "ranges": [
        {
          "events": [
            {
              "introduced": "0"
            },
            {
              "fixed": "1.10.1"
            }
          ],
          "type": "ECOSYSTEM"
        }
      ]
    }
  ],
  "aliases": [
    "CVE-2026-10561"
  ],
  "database_specific": {
    "cwe_ids": [
      "CWE-266",
      "CWE-94",
      "CWE-95"
    ],
    "github_reviewed": true,
    "github_reviewed_at": "2026-10-06T13:38:28Z",
    "nvd_published_at": null,
    "severity": "CRITICAL"
  },
  "details": "### Summary\n\nLangflow\u0027s built-in Python interpreter components \u2014 `PythonREPLComponent` (Python Interpreter) and the legacy `PythonREPLToolComponent` (Python REPL Tool) \u2014 executed arbitrary user- or model-supplied Python code inside flows without effective sandboxing. Because the code ran in-process with the privileges of the Langflow service, any **authenticated** user who could edit and run a flow could achieve remote code execution and, from there, escalate privileges to superuser (e.g. by opening a database session and flipping `is_superuser`) or compromise the host.\n\n**This issue is fixed as of 1.10.1**, with additional hardening through 1.12.3. See *Remediation* below.\n\n### Affected\n\n- **Package:** `langflow` (PyPI), and the underlying `lfx` package that ships the component.\n- **Vulnerable versions:** `\u003c 1.10.1`.\n- **Patched:** `1.10.1` (core fix). Upgrade to `\u003e= 1.12.3` for the complete hardening series.\n\n### Details\n\nThe root cause is **code injection** (CWE-94/CWE-95): the component passed raw input to LangChain\u0027s `PythonREPL`, which is explicitly *not* a security sandbox.\n\nTwo distinct weaknesses existed before 1.10.1:\n\n1. **Unrestricted builtins (default deployments).** `get_globals()` built the `exec` globals from the `global_imports` allow-list but never set `__builtins__`. CPython\u0027s `exec()` then auto-injected the full `builtins` module, leaving `__import__`, `open`, `eval`, `exec` and the whole import machinery reachable regardless of the allow-list \u2014 e.g. `__import__(\"os\").system(...)` or `__import__(\"subprocess\").check_output([...])`. This made the \"only modules in Global Imports can be used\" guarantee false, and it applied even with the **default** configuration (`allow_custom_components=True`).\n\n2. **No server-policy gate (locked-down deployments).** Even a deployment hardened with `allow_custom_components=False` could still run interpreter code, because the components did not consult that policy before executing.\n\nBoth let an authenticated user run the reported PoC, which opens a DB session and sets `is_superuser = True` on their account, or writes to the filesystem / runs OS commands with the service\u0027s privileges.\n\n### PoC (as reported)\n\n1. Authenticate as a normal user.\n2. Create a flow with the `PythonREPLComponent`.\n3. Execute Python that imports Langflow internals and elevates the account:\n\n```python\nimport asyncio\nfrom sqlmodel import select\nfrom langflow.services.database.models.user.model import User\nfrom langflow.services.deps import session_scope\n\nasync def escalate():\n    async with session_scope() as session:\n        stmt = select(User).where(User.username == \u0027testuser\u0027)\n        user = (await session.exec(stmt)).first()\n        if user:\n            user.is_superuser = True\n            session.add(user)\n            await session.commit()\n\nasyncio.run(escalate())\n```\n\n### Impact\n\nAny authenticated user could:\n- Execute arbitrary Python / OS commands with the Langflow service\u0027s privileges (RCE).\n- Escalate their own account to superuser via direct database access.\n- Read/modify data and configuration, and potentially pivot to the underlying host.\n\n### Remediation\n\nUpgrade to **Langflow 1.10.1 or later** (preferably **\u003e= 1.12.3**). The interpreter components were hardened with layered, defense-in-depth controls, applied in `run_python_repl()` before any code is executed:\n\n- **Restricted builtins** \u2014 `get_globals()` injects a curated `safe_builtins()` mapping, removing `__import__`, `eval`, `exec`, `compile`, `open`, `input`, `globals`/`locals`/`vars`, `getattr`/`setattr`, etc. (#13397)\n- **AST validation** \u2014 `validate_code_safety()` rejects inline `import`/`from ... import`, dunder/escape-gadget attribute access (`__class__`, `__subclasses__`, `__globals__`, frame/traceback introspection) and format-string dunder traversal. (#13397)\n- **Server-policy gate** \u2014 `ensure_code_execution_enabled()` refuses to run when `allow_custom_components=False` or `block_code_interpreter_components=True`, and **fails closed** if the settings stack cannot be resolved. (#13700 \u2014 this advisory \u2014 and #14375)\n- **Allow-listed module proxies** \u2014 imported modules are exposed via a proxy that blocks reaching `sys.modules[\"os\"]` through a module\u0027s transitive import graph. (#15198)\n- **Optional hardware isolation** \u2014 configure `LANGFLOW_SANDBOX_BACKEND` to run interpreter code in an isolated microVM instead of in-process. (#14400)\n\nUpstream fix references: #13397, #13700 (carries this GHSA), #14375, #14400, #15198.\n\n### Hardening recommendations for operators\n\n- Keep Langflow updated (\u003e= 1.12.3).\n- For locked-down deployments, set `LANGFLOW_ALLOW_CUSTOM_COMPONENTS=false` (or `LANGFLOW_BLOCK_CODE_INTERPRETER_COMPONENTS=true`) to disable the interpreter entirely.\n- For deployments that must run untrusted code, configure `LANGFLOW_SANDBOX_BACKEND` for microVM isolation.\n- Run the Langflow service as an unprivileged user with least-privilege database credentials.",
  "id": "GHSA-8qpj-27x8-pwpq",
  "modified": "2026-10-06T13:38:28Z",
  "published": "2026-10-06T13:38:28Z",
  "references": [
    {
      "type": "WEB",
      "url": "https://github.com/langflow-ai/langflow/security/advisories/GHSA-8qpj-27x8-pwpq"
    },
    {
      "type": "ADVISORY",
      "url": "https://nvd.nist.gov/vuln/detail/CVE-2026-10561"
    },
    {
      "type": "WEB",
      "url": "https://github.com/langflow-ai/langflow/pull/13700"
    },
    {
      "type": "WEB",
      "url": "https://github.com/langflow-ai/langflow/commit/2754c84aad1306f463db1c3bbb3e8ffbe85da77d"
    },
    {
      "type": "PACKAGE",
      "url": "https://github.com/langflow-ai/langflow"
    },
    {
      "type": "WEB",
      "url": "https://github.com/langflow-ai/langflow/releases/tag/v1.10.1"
    },
    {
      "type": "WEB",
      "url": "https://www.ibm.com/support/pages/node/7277242"
    }
  ],
  "schema_version": "1.4.0",
  "severity": [
    {
      "score": "CVSS:3.1/AV:N/AC:L/PR:L/UI:N/S:C/C:H/I:H/A:H",
      "type": "CVSS_V3"
    }
  ],
  "summary": "Langflow: PythonREPLComponent executes unsandboxed Python code, enabling authenticated RCE and privilege escalation"
}



Log in or create an account to share your comment.




Tags
Taxonomy of the tags.





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.



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.