Key facts
| Fact | Detail |
|---|---|
| Product | IBM Langflow OSS (open-source tool for building AI agents and workflows) |
| Affected versions | 1.0.0 through 1.12.2 |
| Fixed in | 1.12.3 and later (latest: 1.13.0) |
| Severity | 9.8 Critical (CVSS 3.1, IBM) |
| Who can attack | Anyone who can reach the server over the network; no login needed, according to IBM's score |
| Exploited | No attacks reported as of 23 October 2026 |
| CVE published | 6 October 2026 |
What an attacker can do
CVE-2026-104334 is a code injection flaw (the attacker sends input that the server ends up running as code) in Langflow OSS. IBM says a remote attacker could use it to run their own code on your server, with no account and no action from your staff.
The bulletin gives no technical detail: it does not say which part of the product is at fault or how the attack starts. It also leaves out the business impact.
Code running on the server gives the attacker control of the application and of the data it can reach, including stored passwords and keys. If that server can read customer data or holds passwords for other systems, assume those are exposed too. The US cybersecurity agency CISA added notes to the CVE record. They rate the technical impact as "total" and say an attacker can automate the attack.
Who is affected
IBM bulletin 7290694 lists every Langflow OSS release from 1.0.0 through 1.12.2 as affected, without naming a package. This covers the Python packages langflow and langflow-base and the Docker image langflowai/langflow (a ready-to-run package of the application). From 1.12.0 on, langflow-base carries the same version number as langflow. Version 1.12.3 and later are fixed.
IBM names no setting or login role that this flaw needs. Its bulletin covers 25 CVEs and says their impact depends on each install's settings, without saying which ones matter here. Until more detail comes out, treat every affected install as exposed, above all one reachable from outside your company.
On 23 October 2026, the Docker image showed 4,370,570 downloads on Docker Hub. No public report of attacks on this CVE existed that day.
CISA's KEV catalog (the US list of flaws known to be used in real attacks) leaves it out, but already holds six earlier Langflow CVEs, among them CVE-2026-9198, a code injection flaw added on 4 August 2026.
How to check your version
The simplest check asks the running server. Set LANGFLOW_SERVER_URL to your Langflow address, then run:
curl -X GET "$LANGFLOW_SERVER_URL/api/v1/version" -H "accept: application/json"
The version field in the reply gives the langflow package number, for Docker and Python installs alike. On a 1.12.x Python install, that number can differ from the one for langflow-base, so check that package too.
For a pip install, run these in the same Python environment as Langflow, then read each Version: line:
python -m pip show langflow
python -m pip show langflow-base
On Windows, run:
py -m pip show langflow
py -m pip show langflow-base
For an install made with uv, run:
uv pip show langflow
uv pip show langflow-base
The langflow-base line must read 1.12.3 or later. The langflow line alone does not prove the fix.
For Docker, check the image your running container uses:
docker ps --format '{{.Names}} {{.Image}}'
A numbered tag shows the release directly. For the latest tag, or to be sure, use the curl check above, which asks the running server. Anything from 1.0.0 through 1.12.2 needs the update.
What to do now
Back up your data first, as the Langflow team recommends, then update to 1.12.3 or later today. IBM lists no workaround and no signs of compromise.
Langflow 1.13 turns off auto-login (sign-in without a password) by default, as official Docker images already did from 1.11. Once it is off, the old default password langflow stops working. If your team relies on auto-login, set up password sign-in before you update. Deleting or reinstalling the application to regain access can permanently remove your flows (saved workflows).
For uv, update to the latest release or pin a version:
uv pip install langflow -U
uv pip install langflow==1.12.3 langflow-base==1.12.3
langflow 1.12.3 accepts any langflow-base 1.12.x (PyPI dependency langflow-base~=1.12.0), so pinning only langflow does not guarantee the fixed code. langflow-base 1.12.3 in turn requires lfx 1.12.3 or later.
For Docker, check where the container keeps its data. By default, the database, secret key and uploaded files live inside it. Recreating the container deletes them. They survive only with LANGFLOW_DATABASE_URL or LANGFLOW_SAVE_DB_IN_CONFIG_DIR=true, plus LANGFLOW_CONFIG_DIR on a mounted volume (storage outside the container). If not, move the data to a volume first (see the Langflow Docker guide). Then pull langflowai/langflow:1.12.3 or later. Start a new container on the same volume, with LANGFLOW_SUPERUSER_PASSWORD set to a strong password. Images from 1.11 on require it. On an existing install, this variable does not change an admin password that is already set (other than the old default langflow). Change that password in Langflow itself if it is weak. Check the version afterwards.
Get help
Updating closes this flaw. Flaws in your own code get no CVE and no update, and a test of your app is how they are found.
AI agent and chatbot security audit
Can a message, an email or a document make your AI agent act against you? We try, and show you how to stop it.
From AED 35,000, excl. 5% VAT
Sources
- Security Bulletin: Langflow OSS is affected by multiple vulnerabilities IBM
- CVE-2026-104334 record CVE Program
- Known Exploited Vulnerabilities Catalog CISA
- Install Langflow Langflow
- Get started with the Langflow API Langflow
- Deploy Langflow on Docker Langflow
The same update also fixes CVE-2026-93674, CVE-2026-97676, CVE-2026-97678, CVE-2026-97673, CVE-2026-97655, CVE-2026-97679, CVE-2026-88962, CVE-2026-93675, CVE-2026-104335, CVE-2026-93449, CVE-2026-97680, CVE-2026-97674, CVE-2026-103360, CVE-2026-93445, CVE-2026-101331, CVE-2026-93677, CVE-2026-93678, CVE-2026-93443, CVE-2026-93447, CVE-2026-101329, CVE-2026-93448, CVE-2026-97671, CVE-2026-93679.