A critical Langflow vulnerability is being heavily exploited, marking the latest threat to a platform that's become an increasingly popular target.
CVE-2026-0768 is a remote code execution (RCE) flaw in Langflow, a low-code development platform used to design AI agents. The vulnerability, which received a 9.8 CVSS severity score, was initially disclosed in January by Trend Micro's Zero Day Initiative (ZDI).
In a LinkedIn post on Saturday, Caitlin Condon, vice president of security research at VulnCheck, noted that the vendor observed the RCE flaw being exploited on its "canaries," or honeypot systems.
"We've seen 50+ Canary detections for CVE-2026-0768 so far this morning. Adversaries appear to be conducting a mix of reconnaissance and credential harvesting activities," Condon wrote.
At the time of Condon's post, exploitation activity was limited to canaries in the U.K., with attempts coming primarily from IP addresses in Russia. However, attacks have since expanded to include attempts from other countries and canaries elsewhere, once again illustrating the appeal of Langflow to a variety of threat actors.
CVE-2026-0768 Exploitation Broadens
Condon tells Dark Reading that VulnCheck has observed "sustained exploitation" of the Langflow vulnerability since exploitation was first observed on Saturday.
"As of Tuesday, our Canary network has seen exploitation activity from roughly 20 different IPs across more than half a dozen different countries," she says. "We've observed a mix of automated scanning, exploitation for initial access, and post-exploitation activity, including ongoing credential and secrets harvesting (e.g., reading Langflow's secret_key, grepping environment variables for API keys and cloud credentials, searching for SSH keys and .env files), lateral movement, and Langflow source code exfiltration."