SAML Authentication Bypass in Apache Airflow AWS Auth Manager

This repository is a small, focused proof-of-concept for CVE-2026-25604, a Host header injection / origin validation flaw in Apache Airflow's AWS Auth Manager SAML flow. The repo contains only two files: a detailed README explaining the vulnerability, affected versions, attack scenarios, setup, and patch details; and a single Python script, mock_airflow.py, which implements a vulnerable Flask-based mock Airflow service. The exploit capability is authentication bypass via SAML response capture and replay. The PoC reproduces the vulnerable logic by reading the Host header directly from the incoming request in _prepare_request_VULNERABLE(), splitting it into hostname and port, and feeding those values into the SAML service provider configuration. In _init_saml_auth(), the script constructs assertionConsumerService.url as http://<host-from-header>:<port-from-header>/login_callback. Because this ACS URL is derived from attacker-controlled input, a request to /login with a spoofed Host header causes the generated AuthnRequest to direct the IdP to send the signed SAMLResponse to an attacker-controlled endpoint instead of the legitimate Airflow instance. Operational flow in the code: the /login route logs the Host header, initializes SAML auth, and redirects the client to the IdP login URL. The /login_callback route accepts POSTed SAMLResponse data, processes it with python3-saml, and if authentication succeeds, returns a success page and sets a session cookie. This makes the repository more than a pure documentation PoC: it is runnable exploit-demonstration code that shows both the vulnerable request preparation and the replay/processing path. Fingerprintable targets and endpoints include the local Flask routes /, /login, and /login_callback; the dynamically generated ACS URL; the example AWS IAM Identity Center metadata URL; and the example attacker-controlled hosts attacker.com:9090 and evil.com:8080 used in the documentation. The PoC binds on 0.0.0.0 and defaults to port 8080. Overall, the repository's purpose is educational and demonstrative: it does not automate phishing infrastructure or token interception itself, but it clearly shows how a vulnerable Airflow deployment can be coerced into generating a malicious ACS URL and how a captured SAMLResponse can then be replayed to obtain authenticated access.