IBM has published a security advisory detailing multiple critical vulnerabilities in the open-source Langflow project that can be exploited to execute arbitrary code on affected servers.
The advisory, released on August 24, 2026, covers three distinct flaws. CVE-2026-19286 is a code injection vulnerability in the Agent-to-Agent (A2A) public endpoint. The restriction mechanism intended to prevent code execution is not properly implemented, allowing remote attackers to run arbitrary code when A2A is enabled and an agent flow is exposed without requiring authentication.
CVE-2026-19295 affects the evaluation process in flow templates and constitutes an eval injection flaw. An authenticated user can bypass the setting that disables custom components, escalate privileges, and execute arbitrary operating-system-level commands.
The third issue, CVE-2026-18729, results from inadequate control over code generation. Low-privileged authenticated users can circumvent protected input paths and execute arbitrary code directly on the Langflow server.
Updated versions containing fixes for all three vulnerabilities have been released and users are advised to apply the patches immediately.


