Executive brief
LoLLMs, a tool for interfacing with Large Language Models, is vulnerable to a security flaw in how it manages user sessions. The application uses a weak secret key to sign security tokens, which allows an attacker to guess the key and create their own administrative credentials. If exploited, an unauthorized person could take full control of the system, access restricted data, and impersonate administrators.
Technical details
LoLLMs version 2.1.0 and earlier utilizes a weak, predictable secret key for signing JSON Web Tokens (JWT) used in session management. This allows a remote, unauthenticated attacker to perform an offline brute-force attack to recover the secret key. Once recovered, the attacker can forge JWTs with administrative claims, leading to full privilege escalation and unauthorized access to restricted API endpoints. The vulnerability is addressed in version 2.2.0, which implements automatic rotation for weak or missing secret keys.
Affected products
- parisneo lollms < 2.2.0
Timeline
- 2026-04-07: advisory: GitHub Advisory published
- 2026-04-07: disclosed: NVD publication date
- 2026-01-17: patched: Fix committed to repository