Executive brief
AutoGPT, a platform for managing AI agents, contains a flaw in how it processes video files. An attacker can provide a command that forces the system to loop a video an infinite number of times, which quickly fills up the server's disk space. This can crash the service and prevent legitimate users from accessing the platform.
Technical details
A denial-of-service vulnerability exists in AutoGPT's LoopVideoBlock due to uncontrolled resource consumption (CWE-400). The `LoopVideoBlock` component allows users to specify a video file and a number of loops or a duration for processing. However, the application fails to validate or limit the `n_loops` or `duration` parameters. An attacker can provide an extremely high value for these inputs, causing the backend to generate a massive video file that exhausts available disk space. Furthermore, because agent execution can be delayed indefinitely, the resulting disk exhaustion can lead to a persistent DoS state. The issue is fixed in version 0.6.63.
Affected products
- Significant-Gravitas AutoGPT < 0.6.63
Timeline
- 2026-06-11: advisory: GitHub Security Advisory published
- 2026-06-18: disclosed: NVD publication date