Critical Langflow Vulnerability Exploited to Steal OpenAI and AWS Keys

Cybersecurity researchers have reported active exploitation of the critical CVE-2026-0768 vulnerability in Langflow, a popular open-source platform for developing AI applications. The attacks enable unauthenticated arbitrary code execution and subsequent theft of credentials and access keys.

According to VulnCheck, at least 50 exploitation attempts were recorded over the weekend, with the total number of attacks rising to 360. Most traffic originated from IP addresses registered in Russia. Threat actors scan environments to extract environment variables such as OPENAI_API*, AWS_ACCESS*, and AWS_SECRET*, as well as configuration files and command histories.

The CVE-2026-0768 vulnerability affects Langflow versions 1.4.2 and below. It stems from inadequate input validation in the custom component editor, allowing Python code execution with root privileges. As of this report, no public exploits for this vulnerability are known.

This incident is not the first time Langflow vulnerabilities have been exploited this year. Previously, hackers actively used CVE-2026-33017 for code injection and theft of .ENV files, as well as CVE-2026-5027 for arbitrary file writes on servers. Recently, CISA also warned about attacks leveraging CVE-2026-9198 following the release of public PoC exploits.

To mitigate the threat, developers urge Langflow users to immediately update the platform to the latest available version 1.11.6, which addresses all known vulnerabilities.