Executive brief
Langflow, a tool for building AI workflows, is vulnerable to a denial-of-service attack that can take the entire application offline. An attacker can send a specially crafted file upload request that forces the server to consume excessive resources, making it unresponsive to all users. This attack does not require a password or any special access, potentially leading to indefinite service outages.
Technical details
A vulnerability in Langflow's file upload endpoint (/api/v1/files/upload/) allows for unauthenticated denial-of-service via uncontrolled resource consumption (CWE-400). The application fails to validate the multipart form boundary before processing, allowing an attacker to submit a payload containing an extremely large number of hyphens (e.g., 1,000,000+) following the boundary. This causes the server to enter a 'busy' state indefinitely while attempting to parse the malformed data. Additionally, the endpoint lacked authentication checks, allowing remote, unauthenticated attackers to trigger the flaw. Version 1.0.19 fixes this by implementing a 'check_boundary' middleware to validate boundary length and format, and by enforcing authentication and flow-ownership checks.
Affected products
- langflow-ai Langflow < 1.0.19
Timeline
- 2024-09-30: patched: Fix merged in pull request #3923
- 2026-06-19: advisory: GitHub Security Advisory published
- 2026-06-23: disclosed: CVE-2026-55446 published to NVD