High severity · published Aug 6, 2026
CVE-2026-43628
llama.cpp b3978–b9058 Integer Underflow via DRY Sampler
7.8
CVSS
0.2%
EPSS
No
On KEV
Description
llama.cpp builds b3978 through b9058 contain an integer underflow and out-of-bounds read vulnerability in the DRY sampler that allows unauthenticated attackers to trigger a heap buffer underflow by sending a crafted HTTP request with dry_allowed_length set to INT32_MIN to the /v1/completions or /v1/chat/completions endpoints. Attackers can exploit this vulnerability to crash the server with SIGSEGV causing denial of service for all connected users, or corrupt token sampling probabilities by reading garbage values from memory before the allocated buffer.
CVSS vector
CVSS:3.1/AV:L/AC:L/PR:N/UI:R/S:U/C:H/I:H/A:H
Attack vector
Local
Complexity
Low
Privileges
None
User interaction
Required
Confidentiality
High
Integrity
High
Availability
High
Scope
Unchanged
Exploitation
Is it being used?
Exploit prediction (EPSS)0.2% · higher than 14% of CVEs
Exploited in the wildNo evidence yet
CISA KEVNot listed
Ransomware campaign useNot known
Affected products
Which versions
LL
llama.cpp · ggml-org
b3978 through b9058; 0.11.0 through 0.17.1 · source: vendor, NVD
References
NVD
https://nvd.nist.gov/vuln/detail/CVE-2026-43628
CVE.org
https://www.cve.org/CVERecord?id=CVE-2026-43628
Patch
https://github.com/Vladimir-tokarev-cyera/llama-cpp-security-patches
Patch
https://www.vulncheck.com/advisories/llama-cpp-b3978-b9058-integer-underflow-via-dry-sampler
More from llama.cpp
9.8
CVE-2026-34159
llama.cpp: Unauthenticated RCE via GRAPH_COMPUTE buffer=0 bypass in llama.cpp RPC backend
8.2
CVE-2024-58400
llama.cpp before b3701 Out-of-Bounds Read in llama_decode via Token IDs
8.1
CVE-2026-118375
llama.cpp b8841 before b11865 Stack-based Buffer Overflow via ggml-rpc RDMA rdma_recv()
8.1
CVE-2026-107183
llama.cpp before b11393 Use-After-Free via common_chat_peg_mapper chat_parser
8.1
CVE-2026-43632
llama.cpp b7492–b9060 Use-After-Free in Tokenization Endpoints