Executive Summary
In March 2026, a critical unauthenticated remote code execution (RCE) vulnerability, CVE-2026-33017, was disclosed in Langflow, a visual framework for AI agents. The Sysdig Threat Research Team observed active exploitation starting just 20 hours after the advisory, even without a public proof-of-concept. Attackers are targeting the `/api/v1/build_public_tmp/{flow_id}/flow` endpoint to execute arbitrary Python code.
The exploitation followed a clear progression from automated scanning using custom Nuclei templates to advanced credential harvesting and the delivery of second-stage droppers. This rapid weaponization highlights a collapsing 'time-to-exploit' window, where threat actors operationalize vulnerabilities faster than standard patching cycles can accommodate. Security teams must prioritize runtime monitoring for unusual system calls and outbound connections from AI orchestration tools.