A just lately disclosed important safety flaw impacting the open-source Langflow platform has been added to the Recognized Exploited Vulnerabilities (KEV) catalog by the U.S. Cybersecurity and Infrastructure Safety Company (CISA), citing proof of energetic exploitation.
The vulnerability, tracked as CVE-2025-3248, carries a CVSS rating of 9.8 out of a most of 10.0.
“Langflow comprises a lacking authentication vulnerability within the /api/v1/validate/code endpoint that permits a distant, unauthenticated attacker to execute arbitrary code by way of crafted HTTP requests,” CISA mentioned.
Particularly, the endpoint has been discovered to improperly invoke Python’s built-in exec() perform on user-supplied code with out satisfactory authentication or sandboxing, thereby permitting attackers to execute arbitrary instructions on the server.
The shortcoming, which impacts most variations of the favored device, has been addressed in model 1.3.0 launched on March 31, 2025. Horizon3.ai has been credited with discovering and reporting the flaw in February.
In response to the corporate, the vulnerability is “simply exploitable” and permits unauthenticated distant attackers to take management of Langflow servers. A proof-of-concept (PoC) exploit has since been made publicly accessible as of April 9, 2025, by different researchers.
Knowledge from assault floor administration platform Censys exhibits that there are 466 internet-exposed Langflow situations, with a majority of them concentrated in the USA, Germany, Singapore, India, and China.
It is at present not recognized how the vulnerability is being abused in real-world assaults, by whom, and for what objective, though the SANS Expertise Institute mentioned it recorded exploit makes an attempt concentrating on the flaw in opposition to its honeypots. Federal Civilian Govt Department (FCEB) companies have time till Could 26, 2025, to use the fixes.
“CVE-2025-3248 highlights the dangers of executing dynamic code with out safe authentication and sandboxing measures,” Zscaler famous final month. “This vulnerability serves as a important reminder for organizations to strategy code-validation options with warning, notably in purposes uncovered to the web.”
Replace
Assault floor administration platform Censys mentioned it has noticed 1,156 uncovered Langflow servers on-line, practically half of that are situated in the USA. Of those, 360 servers look like working a susceptible model. As many as 509 servers didn’t promote a model.
