- CVSS
- MEDIUM · 6.6v3.1CVSS:3.1/AV:L/AC:L/PR:N/UI:R/S:U/C:L/I:L/A:H
- Published
- 2026-10-27
- Weakness
- CWE-190
- Source
- nvd.nist.gov/vuln/detail/CVE-2026-116829
Description
llama.cpp b10398 before b11301 contains an integer overflow vulnerability in gguf_init_from_reader that allows attackers to bypass tensor element-count validation by supplying shapes whose product wraps to zero. Attackers can craft GGUF files with enormous tensor extents that overflow ggml_nbytes() and strides, causing out-of-bounds memory access when buffers are sized from wrapped byte counts.
References
- https://github.com/ggml-org/llama.cpp/blob/v0.5.0/ggml/src/gguf.cpp#L701
- https://github.com/ggml-org/llama.cpp/commit/2149c00f4442dc59302e134a02e4c99d5f7ed9fc
- https://github.com/ggml-org/llama.cpp/issues/29383
- https://github.com/ggml-org/llama.cpp/pull/29384
- https://github.com/ggml-org/llama.cpp/releases/tag/b11301
- https://www.vulncheck.com/advisories/llama-cpp-b10398-before-b11301-integer-overflow-in-gguf-init-from-reader
How GTK Cyber trains on this
AI security training at GTK Cyber covers the LLM and ML-pipeline vulnerability classes that vulnerabilities like CVE-2026-116829 fall into. Our hands-on courses are taught by Charles Givre and other practitioners who break and defend production AI systems.