Skip to main content
Vulnerability Database/CVE-2026-107366

CVE-2026-107366: IBM Langflow OSS Auth Bypass Vulnerability

CVE-2026-107366 is an authentication bypass flaw in IBM Langflow OSS allowing authenticated users to manipulate other users' messages and hijack conversations. This post covers technical details, affected versions, impact, and mitigation steps.

Published:

Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.

Sources: National Vulnerability Database, CISA KEV, Exploit-DB

CVE-2026-107366 Overview

CVE-2026-107366 is a broken access control vulnerability in IBM Langflow OSS versions 1.0.0 through 1.12.4. The flaw affects three message endpoints that fail to perform ownership verification on incoming requests. Any authenticated user can bulk-delete another user's messages, overwrite message content to inject malicious prompts, or hijack a conversation session by reassigning its session identifier. The weakness is tracked under CWE-639: Authorization Bypass Through User-Controlled Key.

Critical Impact

Authenticated attackers can tamper with other users' conversations, inject prompt-injection payloads, and hijack sessions in multi-tenant Langflow deployments used for building LLM-powered applications.

Affected Products

  • IBM Langflow OSS 1.0.0 through 1.12.4
  • Deployments exposing the vulnerable message endpoints to multi-user environments
  • Downstream applications built on affected Langflow versions

Discovery Timeline

  • 2026-10-20 - CVE-2026-107366 published to NVD
  • 2026-10-21 - Last updated in NVD database

Technical Details for CVE-2026-107366

Vulnerability Analysis

IBM Langflow OSS exposes message management endpoints that handle conversation history for LLM workflows. Three of those endpoints accept message identifiers and session identifiers directly from the request without validating that the authenticated caller owns the referenced objects. The server trusts the client-supplied key and performs the requested delete, update, or reassignment action against any matching record.

An authenticated attacker can enumerate or guess message and session identifiers belonging to other tenants. They can then bulk-delete those messages, destroying conversation history and causing an integrity and availability impact. They can also rewrite stored message content, enabling stored prompt-injection attacks that execute when the victim replays the conversation through an LLM. Reassigning a session ID transfers an entire conversation to an attacker-controlled context, delivering a full session hijack.

Root Cause

The root cause is missing ownership checks in the handlers for the three message endpoints. The code paths authenticate the user but never compare the user's identity against the owner of the targeted message or session row. This pattern maps directly to CWE-639, where authorization depends on a user-controlled key rather than a server-side ownership lookup.

Attack Vector

Exploitation requires network access and a valid low-privileged account on the Langflow instance. No user interaction is required. The attacker issues direct HTTP requests to the vulnerable endpoints while supplying identifiers that reference other users' data. Because Langflow is commonly used to build and expose LLM pipelines, successful prompt overwrite can propagate malicious instructions into downstream model calls and connected tools.

See the IBM Support Page for vendor technical details.

Detect Identity Attacks the Moment Authentication Fails

Singularity™ Identity surfaces credential misuse, lateral movement, and privilege escalation in real time, so you can contain the breach before it spreads.

Stop Identity Attacks

Detection Methods for CVE-2026-107366

Indicators of Compromise

  • Unexpected bulk DELETE requests against Langflow message endpoints from a single authenticated session
  • PATCH or PUT requests that modify messages whose owner does not match the authenticated principal
  • Session identifier reassignments that move conversations between distinct user accounts
  • Prompt or message content containing injection markers such as instructions to ignore prior context

Detection Strategies

  • Enable verbose application logging on Langflow message endpoints and record user_id, message_id, and session_id for every mutation
  • Correlate the authenticated user identity against the message owner recorded in the database to surface cross-tenant access
  • Alert on high-volume delete or update operations issued by a single account within short time windows

Monitoring Recommendations

  • Forward Langflow access and audit logs to a centralized analytics platform for cross-user correlation
  • Baseline normal per-user message volume and alert on deviations consistent with enumeration or bulk tampering
  • Review LLM output logs for signs of injected instructions originating from modified message history

How to Mitigate CVE-2026-107366

Immediate Actions Required

  • Upgrade IBM Langflow OSS to a fixed release as directed by the IBM Support Page
  • Restrict network exposure of Langflow instances to trusted users and networks until patching completes
  • Rotate active session identifiers and audit recent message mutations for unauthorized changes

Patch Information

IBM has published remediation guidance on its support portal. Administrators running IBM Langflow OSS 1.0.0 through 1.12.4 should apply the vendor-supplied fix referenced in the IBM Support Page. The patched build enforces server-side ownership checks on the affected message endpoints.

Workarounds

  • Place Langflow behind an authenticating reverse proxy that enforces per-tenant isolation at the request layer
  • Disable or network-restrict the vulnerable message endpoints where feasible until the patch is deployed
  • Limit account provisioning to trusted users and avoid exposing shared Langflow instances to untrusted authenticated populations
bash
# Example: restrict Langflow message endpoints at an Nginx reverse proxy
location ~ ^/api/v1/(messages|monitor/messages) {
    allow 10.0.0.0/8;
    deny all;
    proxy_pass http://langflow_upstream;
}

Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.

Sources: National Vulnerability Database, CISA KEV, Exploit-DB

Experience the Most Advanced Cybersecurity Platform

See how the world’s most intelligent, autonomous cybersecurity platform can protect your organization today and into the future.