The flaw lies in CVAT’s automatic annotation endpoint, where a user with write access to a specific job can send a batch annotation request that pairs a task ID with a different job ID. Because the task ID controls a single active request slot, misaligned IDs let the attacker block automatic annotation for a target task, causing a denial of service for that task’s annotation workflow. This is a logic error that permits manipulation of request routing (CWE‑1288).

The vulnerability affects CVAT version 2.17.0 through 2.72.0, provided by cvat‑ai as an open‑source tool for video and image annotation. Any installation running one of these releases is potentially impacted until the patch in the next release is applied.

The CVSS score is 5.3, indicating moderate severity, while the EPSS score is below 1 %, showing very low exploitation probability. The flaw is not listed in CISA KEV, suggesting no publicly known exploits. Exploitation requires authenticated access and knowledge of a valid task ID that is currently handling an automatic annotation request, limiting the attack surface. Nevertheless, an internal actor or compromised user with job‑write privileges could interrupt the annotation process for a particular task, leading to operational disruption.