AutoGPT Platform EUVD-2025-209827

Exploitation requires Docker container deployment mode of AutoGPT Platform - the vulnerability does not apply to non-containerized deployments where Docker's log capture is absent. High sustained user access volume or deliberately verbose workflow execution is necessary to trigger meaningful disk exhaustion; SSVC classifies this as not automatable, confirming a single request does not cause the condition. The CVSS AV:L (local attack vector) designation is in tension with the description's reference to user access count as the trigger mechanism - if network-sourced traffic drives log growth, the practical exploitation surface may be broader than AV:L implies, and this should be confirmed with the vendor. No confidentiality or integrity impact is possible under any exploitation path. Authentication requirements for triggering log growth depend on whether the AutoGPT platform endpoint itself requires authentication, which is not specified in the CVE data.

CVSS 4.0 scores this at 5.1 (Medium) with an attack vector of Local (AV:L), low attack complexity, no privileges required, and no user interaction - but impact is limited to local availability (VA:L) with no confidentiality or integrity consequence. There is a notable tension between AV:L and the description's framing of 'number of user accesses' as the trigger: if network-facing user traffic drives log growth, the effective attack vector may be network-adjacent rather than strictly local. This discrepancy should be verified with Significant Gravitas before downgrading risk. EPSS at 0.01% (3rd percentile) and SSVC Automatable=No confirm that a single crafted request does not trigger exploitation - sustained volume is required. SSVC Technical Impact is partial, consistent with VA:L. Overall, this is a low operational priority for low-traffic or properly log-rotated deployments, but can represent a real availability risk for high-traffic AutoGPT instances running in Docker without external log management.

An operator runs AutoGPT Platform in Docker container mode exposed to many concurrent users executing complex AI workflows. Over hours or days, the platform's activity, debug, and error log files grow without bound as Docker captures all stdout/stderr, eventually filling the host disk volume. Once disk capacity is exhausted, the container and all co-located services fail with ENOSPC errors, taking the entire platform offline. No public exploit code exists - this requires sustained usage volume rather than a single crafted payload.