Executive brief
Feast is an open-source feature store used to manage and serve machine learning features. The platform fails to verify JWT token signatures before accepting user identity claims, allowing attackers to forge authentication tokens and gain full read/write access to all data, feature definitions, and configuration on the server.
Technical details
The vulnerability is an authentication bypass in Feast through version 0.66.0 caused by insufficient JWT token validation. The server accepts user identity claims from JWT tokens without cryptographically verifying the token signature, allowing an attacker to present a forged or unsigned token with hardcoded claim values to impersonate any user. This bypasses all role-based access control (RBAC) checks. No authentication is required beyond sending a malicious JWT token; the attack is trivial to execute from the network. Successful exploitation grants unchecked read and write access to all entities, feature views, data sources, and permission policies, including the ability to modify or steal sensitive ML training data. Patches are expected in versions 0.66.1 and later.
Affected products
- Feast Feast through 0.66.0
Timeline
- 2026-09-16: disclosed