Security Bulletin

CVE-2026-9198: Critical Langflow RCE Under Active Exploitation

3 min read Updated

CISA confirmed active exploitation of CVE-2026-9198, adding it to the Known Exploited Vulnerabilities catalog on August 4, 2026.

The vulnerability lets an unauthenticated attacker chain two API endpoints to gain full remote code execution on a default Langflow deployment. IBM, which maintains Langflow, disclosed the vulnerability on July 17, 2026, and shipped a fix the same day. The company is now urging all Langflow OSS users to apply that fix immediately, given the confirmed exploitation activity.

CVE-2026-9198: How the Endpoint Chain Leads to Full RCE

Severity
Critical
CVSS v3.1 Base Score
9.8
Exploit Available in Public
Yes
Exploitation Complexity
Low
Attack Vector
CVSS:3.1/AV:N/AC:L/PR:N/UI:N/S:U/C:H/I:H/A:H

CVE-2026-9198 is a code injection vulnerability, tracked under CWE-94, in IBM Langflow OSS versions 1.0.0 through 1.10.0. The root cause traces to an insecure default configuration combined with unsafe use of dynamic code evaluation. The /api/v1/auto_login endpoint fails to enforce authentication and isn’t bound to loopback, so it mints SUPERUSER tokens for any network caller rather than restricting access to the local host. The /api/v1/validate/code endpoint then trusts that authenticated input and evaluates attacker-supplied Python directly through exec().

Chained together, an unauthenticated attacker can mint a superuser token from the first endpoint, then hand arbitrary Python code to the second endpoint to run with full administrative privileges. On a default Langflow deployment, that is complete remote code execution with no login, no user interaction, and no prior access required.

CVE-2026-9198: Affected Versions

Detail Value
Affected versions Langflow OSS 1.0.0 – 1.10.0
Disclosure date July 17, 2026
Fixed in 1.10.1Patched same day
Current stable release 1.11.2

Anyone running a self-hosted or containerized Langflow instance should confirm their build version against IBM’s support.

CVE-2026-9198: Why This Vulnerability Matters

As an AI workflow and agent-orchestration tool, Langflow typically holds model-provider API keys, database credentials, connector tokens, and access to whatever files or systems its flows are wired into. An attacker who achieves RCE on a Langflow host gets a path to every credential and connected system that Langflow was trusted with.

🚨 CISA KEV Confirmation: CVE-2026-9198 was added to CISA’s Known Exploited Vulnerabilities catalog on August 4, 2026. Confirmed exploitation activity means any internet-facing Langflow instance should be treated as a live target.

Public exploit code is already circulating, with reports pointing to exploit-development artifacts and staging infrastructure consistent with active testing and weaponization, separate from any confirmed victim compromise. A related campaign has already been tied to an AI-enabled autonomous hacking operation that pivoted between multiple Langflow and adjacent tooling vulnerabilities when initial attempts were blocked.

The threat actor used an AI agent to independently research alternate entry points. That combination, a low-effort exploit chain, public PoC circulation, and demonstrated attacker automation, makes any internet-facing Langflow instance a live target.

CVE-2026-9198: Mitigation Recommendations

Since this vulnerability is already being actively exploited, the priority is closing the exploit path immediately rather than treating it as a routine patch cycle item. Follow these steps to reduce exposure:

Upgrade Langflow to version 1.10.1 or later

Redeploy any containerized instances using the updated image rather than patching in place, since older container builds can silently persist the vulnerable code.

Remove Langflow from direct internet exposure

Place it behind an authenticating reverse proxy or VPN, so the vulnerable endpoints are never reachable by unauthenticated network callers in the first place.

Rotate all secrets and credentials

Rotate API keys and model-provider credentials stored in Langflow flows on any host that was internet-facing before patching, since a compromised instance may have already exposed them.

Audit affected hosts for prior exploitation

Look for unfamiliar cron entries, new SSH keys, and unexpected outbound connections.

Verify the fix took effect

Confirm /api/v1/auto_login no longer issues tokens to unauthenticated callers and is bound to loopback after upgrading.

Track exploitation activity going forward

Monitor EPSS movement and exploitation reporting for this CVE alongside routine scans of internet-facing Langflow assets, since attacker tooling is still evolving.

AppTrana WAAP Coverage for CVE-2026-9198

AppTrana blocks exploitation attempts targeting this vulnerability, giving customers running Langflow behind AppTrana protection while patching is completed. AppTrana WAAP inspects incoming requests to internet-facing applications and blocks traffic that matches known malicious patterns, including attempts to abuse authentication and code-execution endpoints.

Protect Langflow with AppTrana WAAP

Block exploitation of CVE-2026-9198 while your team completes the upgrade to 1.10.1+.

See AppTrana in Action

Deepak
Deepak Kumar Choudhary

Deepak Choudhary is an adept Security Researcher at Indusface. He specializes in developing detection logic and signatures to identify various security vulnerabilities, including 0-day vulnerabilities, making him a frontline defender of digital environments.Driven by his passion for cyber defense, He continuously seeks to expand his knowledge of security concepts. He eagerly tackles the task of resolving vulnerable systems on platforms such as TryHackMe and HackTheBox, sharpening his skills in real-world scenarios. Through his expertise and dedication, he is committed to fortifying digital landscapes, ensuring a safer online experience for users and organizations alike.