vLLM from 0.19.0 before 0.31.0 contains a resource exhaustion vulnerability that allows inference clients to bypass the server frame limit by setting media_io_kwargs.video.num_frames in chat completion requests. Attackers can supply num_frames=-1 or a large value with many base64 JPEG frames, forcing VideoMediaIO to decode all frames and np.stack them, causing memory exhaustion.
{
"cna_assigner": "VulnCheck",
"cwe_ids": [
"CWE-770"
],
"osv_generated_from": "https://github.com/CVEProject/cvelistV5/tree/main/cves/2026/126xxx/CVE-2026-126399.json"
}{
"extracted_events": [
{
"range": [
{
"introduced": "0.19.0"
},
{
"fixed": "0.31.0"
}
],
"source": "AFFECTED_FIELD",
"strategy": "StandardRange"
},
{
"range": [
{
"introduced": "0.19.0"
},
{
"fixed": "0.31.0"
}
],
"source": "DESCRIPTION"
}
]
}