Attackers Exploiting Critical Langflow Flaw to Steal OpenAI and AWS Keys

Overview

• VulnCheck reported Tuesday that its UK honeypots saw an initial wave of about 50 exploitation attempts over a weekend that grew to roughly 360 observed attacks targeting the Langflow code-validator bug.
• The flaw, tracked as CVE-2026-0768 with a CVSS score of 9.8, exists in Langflow’s custom component code validator and allows unauthenticated execution of arbitrary Python code as root on versions up to 1.4.2.
• Observed attacker activity focuses on reconnaissance and secret harvesting by querying environment variables and reading files to capture LANGFLOW_SUPERUSER tokens, OPENAI_API keys, AWS_ACCESS and AWS_SECRET values, cached Langflow secret_key files, and .ssh data.
• There are no known public proof-of-concept exploits for CVE-2026-0768, but Langflow flaws have been weaponized quickly in 2026 so defenders are urged to apply patches now and upgrade to Langflow 1.11.6 or later.
• The incidents form part of a wider rise in adversary interest in Langflow this year, with multiple additional CVEs observed exploited in the wild and thousands of attacks against other vulnerable instances that can lead to cloud abuse and downstream intrusions.