CVE-2026-111392
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.
- CVSS 4.0
- 6.3
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 - CVSS 3.1
- 4.8
CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:L/I:L/A:N - EPSS
- 0.2% chance of exploitation in the next 30 days; higher than 10% of CVEs (scores of 2026-10-25)
- CWE
- CWE-345
- Published
- Modified
- Record from
- the NVD, synced
References
- https://github.com/sgl-project/sglang/blob/v0.5.21/python/sglang/srt/managers/io_struct.py#L205-L215
- 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/issues/44641
- https://github.com/sgl-project/sglang/pull/44653
- https://www.vulncheck.com/advisories/sglang-through-0.5.21-kv-cache-poisoning-via-mm-hashes-in-generate