CVE-2026-45345: Open WebUI improper authorization in model update function

Executive brief

Open WebUI is a user interface for managing and interacting with AI models. A security flaw allows any logged-in user to modify AI models created by other users, even if those models are marked as private. This could allow an attacker to change model settings or grant themselves unauthorized access to restricted models, potentially compromising the integrity of AI operations.

Technical details

An improper authorization vulnerability (CWE-285) exists in Open WebUI versions up to 0.5.6. The application fails to perform adequate ownership or permission checks when a user submits a POST request to the `/api/v1/models/model/update` endpoint. By providing the unique identifier of a target model in the request, an authenticated attacker can modify the model's configuration, metadata, and access control lists, regardless of whether the model is set to 'Private'. This allows for unauthorized modification of model parameters and the escalation of access privileges. The issue is addressed in version 0.5.7.

Affected products

• Open WebUI Open WebUI <= 0.5.6

Timeline

• 2026-05-14: advisory: GitHub Advisory GHSA-gm54-m39w-grjp published
• 2026-05-14: patched: Fixed in version 0.5.7

References

• https://api.github.com/users/simioni87
• https://github.com/simioni87
• https://api.github.com/users/simioni87/gists%7B/gist_id%7D
• https://api.github.com/users/simioni87/repos
• https://avatars.githubusercontent.com/u/37290182?v=4
• https://api.github.com/users/simioni87/events%7B/privacy%7D