Airflow CVE-2025-57735

Exploitation requires the attacker to first obtain a valid JWT token through interception, which depends on: deployment of Airflow over HTTP rather than HTTPS (enables passive network sniffing), access to the same network segment as authenticated users (for MITM or ARP spoofing attacks), presence of XSS vulnerabilities in Airflow or co-hosted applications (enables JavaScript-based token theft from browser storage), or physical/remote access to client devices where tokens are cached in browser memory or disk. The vulnerability itself does NOT create a token bypass - attackers MUST possess a legitimately issued token. Exploitation is LIMITED by: the JWT's built-in expiration time (default varies by Airflow configuration, typically 30-120 minutes), requirement for the user to have successfully authenticated before logout (cannot exploit tokens that were never issued), and network positioning to intercept tokens (purely remote attackers without network access cannot exploit without a separate token-stealing vulnerability). Organizations using HTTPS exclusively and network segmentation significantly reduce the attack surface, though insider threats or device compromise remain viable vectors.

The CVSS 9.1 (Critical) score reflects a theoretical worst-case scenario: network-accessible (AV:N), low complexity (AC:L), no authentication required (PR:N), no user interaction (UI:N), with high confidentiality and integrity impact. However, this rating is MISLEADING because exploitation requires the attacker to first INTERCEPT a valid JWT token - the vulnerability does NOT create tokens out of thin air. Real-world risk depends heavily on whether an attacker can obtain tokens via network sniffing (requires non-HTTPS deployment, rare in production), XSS attacks (separate vulnerability), or compromised client devices. The EPSS score of 0.01% (3rd percentile) and SSVC assessment (exploitation: none, automatable: yes, technical impact: partial) provide crucial context: despite the high CVSS, there is no evidence of active exploitation, and the attack requires a prerequisite step (token interception) not reflected in the CVSS vector's PR:N designation. The PR:N rating appears to assume the attacker already possesses the token, but obtaining that token IS a privilege requirement in practical terms. For organizations using HTTPS (industry standard), not exposing Airflow to untrusted networks, and without concurrent XSS vulnerabilities, real-world exploitability is significantly lower than the CVSS suggests. Organizations with Airflow exposed to public networks, using HTTP, or with shared client devices face elevated risk. CRITICAL DATA CONFLICT: CVSS PR:N contradicts the description's implicit requirement that attackers must first intercept valid tokens - verify whether 'no privileges required' assumes token possession or refers to a separate bypass vector.

An attacker positions themselves on the same network as an Airflow administrator (via compromised WiFi, man-in-the-middle on corporate network, or malicious browser extension on a shared workstation) and captures the JWT token from an authenticated session through packet sniffing of HTTP traffic or extraction from browser local storage. After the administrator logs out of Airflow, believing the session is terminated, the attacker replays the captured JWT in API requests to Airflow's REST endpoints. The token remains valid until its expiration timestamp, allowing the attacker to trigger workflow DAGs, access sensitive connection credentials stored in Airflow, or modify pipeline configurations. No public proof-of-concept code identified, but token replay is a well-understood attack technique requiring only basic HTTP client tools like curl or Postman.