IBM patched critical vulnerabilities in Langflow, an open-source framework for building AI agents, LLM-powered workflows, and retrieval-augmented generation (RAG) systems, that could allow attackers to bypass authentication, leak API credentials, and execute arbitrary code. The latest patches fix four vulnerabilities, two rated critical with 9.8 CVSS score. CVE-2026-93680 stems from an incomplete remediation of a previous auto-login security issue and could let attackers bypass security restrictions and access features normally requiring authentication. CVE-2026-110186 allows unauthenticated remote attackers to execute arbitrary code through Model Context Protocol (MCP) tool enumeration. CVE-2026-110193, rated high severity, also allows code and command execution on the server due to an issue in a third-party library used by the Agentic Assistant, but exploiting it requires authentication with at least a low-privileged account. CVE-2026-110192 is an information disclosure issue due to insufficiently protected API credentials, rated 6.5 on the CVSS scale. The flaws affect Langflow OSS versions 1.0.0 to 1.12.8 and were fixed in 1.13.0, available from PyPI and as a Docker image. IBM advised backing up the database, secret key, and uploaded files before upgrading, and preparing the existing account for password sign-in if automatic login is used. IBM also released patches for three