Do you have an incident?

Our S.O.S. line:

+49 89 262 025954

Our team of experts is ready to assist your organization in the event of a cyberattack.

details

From Auto-Login to Full RCE: Inside the IBM Langflow CVE

Penetration Testing WhiteHat todayAugust 21, 2026

Background

AI has been a hotspot for everything recently and that’s true for attacks too proven by the recently discovered and also exploited IBM Langflow OSS (versions 1.0.0 through 1.10.0) vulnerability CVE-2026-9198, our #CVEoftheWeek.

The issue was addressed in the release of version 1.10.1 on 24th June, which contained many security updates, but known Proof-of-Concepts (PoC) and the fact that the vulnerability was added to the CISA Known Exploited Vulnerabilites Catalog (KEV) has significantly increased the risk of those unaware, also increasing it’s Exploit Prediction Scoring System (EPSS) score to an unusually high 17%. What makes the vulnerability particularly troublesome is that it does not require any authentication and lead to a Remote Code Execution (RCE).

The vulnerability chains two existing flaws. The first one is that a request “GET /api/v1/auto_login” generates a SUPERUSER JWT to any caller. Worth noting that this only happens with default configuration when AUTO_LOGIN is enabled. Additionally, if cross-origin resource sharing (CORS) is not configured properly, it may allow the JWT tokens to be exposed to unintended origins which also increases the risk of unauthorized access. The second one is that a request “POST /api/v1/validate/code” passes user-supplied Python to exec() with no sandbox.

These two separately could already cause troubles like information disclosure by unauthenticated access to a read-only API through SUPERUSER auto_login and privilege escalation for authenticated user through the Python exec() with no sandbox but together they lead to an unauthenticated RCE.

Defender should focus on:

  • Updating to the latest version by IBM recommendation
  • In case of update is not possible for some reason, restrict access to the IBM Langflow instances
  • Review logs and create detections for unexpected POST requests to the Langflow API

For more information you should check IBM’s own report on the issue and if you are interested, take a look at a POC:


Do you want to be protected from attacks like the above? See our defensive services here: Managed Security – White Hat IT Security

Written by: WhiteHat

Tagged as: , , .

Previous post

Penetration Testing WhiteHat / August 14, 2026

Bad Build: RCE Hits TeamCity

This week’s CVE of the Week is CVE-2026-63077, an unauthenticated remote code execution vulnerability in JetBrains TeamCity. It is a deserialization issue in the popular CI/CD service, which leads to executing system commands in the operating system with recently published [...]


Similar posts

Incident Response WhiteHat / September 4, 2026

Metabase Under Attack: A Critical SQL Injection Flaw

This CVE of the Week is ‘Metabase SQL Injection Vulnerability’ (CVE-2026-72898) with an impressive score of 10.0. The exploit utilizes one of the oldest tool in the hackers’ toolbox: #SQLi. In the popular Open-source Business Intelligence tool, Metabase. Metabase enables users to query databases, create visualizations and build dashboard without writing any SQL code with ...

Read more trending_flat