CVE-2026-111392

medium

Description

SGLang through 0.5.21 trusts caller-supplied mm_hashes values as the multimodal prefix-cache key, allowing remote attackers to poison cached KV entries across requests. Attackers can send a different image under a known or predicted hash via /generate to disclose another request's image content or corrupt victims' responses.

References

https://www.vulncheck.com/advisories/sglang-through-0.5.21-kv-cache-poisoning-via-mm-hashes-in-generate

https://github.com/sgl-project/sglang/pull/44653

https://github.com/sgl-project/sglang/issues/44641

https://github.com/sgl-project/sglang/blob/v0.5.21/python/sglang/srt/managers/tokenizer_manager.py#L1183-L1215

https://github.com/sgl-project/sglang/blob/v0.5.21/python/sglang/srt/managers/io_struct.py#L205-L215

https://euvd.enisa.europa.eu/vulnerability/EUVD-2026-103024

Details

Source: Mitre, NVD

Published: 2026-10-16

Updated: 2026-10-16

Risk Information

CVSS v2

Base Score: 4

Vector: CVSS2#AV:N/AC:H/Au:N/C:P/I:P/A:N

Severity: Medium

CVSS v3

Base Score: 4.8

Vector: CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:L/I:L/A:N

Severity: Medium

CVSS v4

Base Score: 6.3

Vector: CVSS:4.0/AV:N/AC:H/AT:P/PR:N/UI:N/VC:L/VI:L/VA:N/SC:N/SI:N/SA:N

Severity: Medium

EPSS

EPSS: 0.00206