Executive brief
AutoGPT, a platform for managing autonomous AI agents, is vulnerable to a denial-of-service attack. A user can create a workflow that repeatedly downloads large media files without deleting them, eventually filling up the server's disk space. This can cause the entire platform to crash or become unavailable, disrupting business operations and AI agent tasks.
Technical details
A denial-of-service vulnerability exists in AutoGPT's `AddAudioToVideoBlock` and `StepThroughItemsBlock` components. The `AddAudioToVideoBlock` downloads and stores video and audio files in a temporary directory but fails to delete them before all nodes complete execution, and it lacks disk space consumption limits. Furthermore, the `StepThroughItemsBlock` allows for unlimited iterations of media processing tasks. An attacker with low privileges can exploit these behaviors via the network to cause uncontrolled disk space exhaustion. The issue is exacerbated by the fact that processed output files are also stored in the working directory without immediate cleanup. This vulnerability is patched 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