CVE-2026-111392: medium-severity vulnerability in sgl-project sglang
SGLang through 0.5.21 KV Cache Poisoning via mm_hashes in /generate
Published
E-mail alert when it lands in KEV, gets a public PoC or its score goes up.
13Vexday Risk Score
No sign of exploitation. No public exploitation artifact known so far.
ssvc Trackcvss 6.3epss 0.2%
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/AV:N/AC:H/AT:P/PR:N/UI:N/VC:L/VI:L/VA:N/SC:N/SI:N/SA:N
Affected products
sgl-project · sglangRelated CVEs — sgl-project sglang
In the same product, most dangerous first.
CVE-2026-93688HIGHSGLang through 0.5.19 Unbounded Memory Allocation via bootstrap_roomEPSS 0.7%CVE-2026-93838HIGHSGLang through 0.5.20 Unbounded Memory Allocation via STAGING_REQ chunk_idxEPSS 0.7%CVE-2026-102634HIGHSGLang through 0.5.20 Denial of Service via Duplicate bootstrap_roomEPSS 0.6%CVE-2026-109226HIGHSGLang through 0.5.21 Audio Decompression Amplification via load_audioEPSS 0.6%CVE-2026-109229HIGHSGLang through 0.5.21 Decompression Bomb DoS via NanoNemotronVLImageProcessorEPSS 0.5%CVE-2026-109227HIGHSGLang through 0.5.21 Memory Exhaustion DoS via Qwen-VL Video PreprocessingEPSS 0.5%
References
https://github.com/sgl-project/sglang/blob/v0.5.21/python/sglang/srt/managers/io_struct.py#L205-L215https://github.com/sgl-project/sglang/blob/v0.5.21/python/sglang/srt/managers/tokenizer_manager.py#L1183-L1215https://github.com/sgl-project/sglang/issues/44641https://github.com/sgl-project/sglang/pull/44653https://www.vulncheck.com/advisories/sglang-through-0.5.21-kv-cache-poisoning-via-mm-hashes-in-generate