Executive brief
vantage6 is a framework for privacy-preserving federated learning. A security flaw allows an algorithm developer to modify the details of an algorithm submitted by a different developer while it is still under review. This could lead to an unauthorized container image being approved and executed across the network, potentially compromising the integrity of the data processing tasks.
Technical details
An incorrect authorization vulnerability (CWE-863) exists in vantage6 where the edit permission for algorithms lacks an ownership check. A remote attacker with developer-level privileges can modify the metadata of algorithms submitted by other users, specifically during the 'pending' or 'under review' phase. By altering the container image or image tag before final approval, an attacker can ensure that nodes execute a different, potentially malicious image than the one originally reviewed. This vulnerability affects all versions up to and including 5.0.2, and no patch is currently available.
Affected products
- vantage6 vantage6 <= 5.0.2
Timeline
- 2026-07-23: disclosed
- 2026-07-24: advisory: GitHub Advisory published