Executive brief
AutoGPT is an automation platform used to create and manage autonomous AI agents. A flaw in the platform's billing logic allows users to execute AI tasks and workflows without consuming their allocated credits. This enables unauthorized free use of expensive AI resources (like GPT-4), potentially leading to significant financial loss for the service provider and exhaustion of system resources.
Technical details
A business logic vulnerability exists in the AutoGPT platform where the credit balance check is improperly enforced. While the platform checks for sufficient credits during full graph executions in 'manager.py', it fails to perform this check when individual blocks are executed directly via the 'POST /api/blocks/{block_id}/execute' endpoint. An authenticated attacker can bypass the billing mechanism to run AI/LLM blocks (such as GPT-4 or Claude) regardless of their actual credit balance. This is classified as an improper enforcement of behavioral workflow (CWE-841) and resource allocation without limits (CWE-770). The issue is resolved in version 0.6.59 by implementing credit verification at the API layer.
Affected products
- Significant-Gravitas AutoGPT (autogpt-platform-beta) < 0.6.59
Timeline
- 2026-05-11: advisory: GitHub Security Advisory published by vendor
- 2026-05-28: disclosed: CVE published to NVD
- 2026-05-28: patched: Fix released in version 0.6.59