# CVE-2026-116467: ggml-org llama.cpp reachable assertion in Mirostat Sampler

Source: https://junglewise.ai/threats/cve-2026-116467-a-security-flaw-has-been-discovered-in-ggml-org-llama-cpp-up-to  
Publisher: Junglewise Threat Intelligence, https://junglewise.ai/threats  
Published: 2026-10-26. Updated: 2026-10-26  
License: CC BY 4.0 with a link back

CVE-2026-116467 · Severity: medium · CVSS 5.3 · Published 2026-10-26

llama.cpp is a C/C++ library that performs inference for large language models. A remote attacker can trigger a crash by supplying non-finite numerical values to the Mirostat sampling component, disrupting service availability for applications using this library.

Technologies: [Ggml-Org Llama.Cpp](https://junglewise.ai/threats/technologies/ggml-org-llama-cpp). Vendors: [Ggml-Org](https://junglewise.ai/threats/vendors/ggml-org).

## Quick answers

**Is CVE-2026-116467 exploited in the wild?** No. Junglewise Threat Intelligence has no report of CVE-2026-116467 being exploited in the wild.

**How severe is CVE-2026-116467?** Medium severity, with a CVSS score of 5.3, published 26 October 2026.

**Which products and versions are affected?** Ggml-Org Llama.Cpp up to 0.7.0.

**Is there a fix for CVE-2026-116467?** No fixed version is recorded yet. Check the vendor's advisory in the references below. If you run it, Junglewise drafts a mitigation that fits your setup and applies it once you approve, with senior security engineers behind it.

**How is CVE-2026-116467 attacked?** It is attacked over the network, in a software library.

**Am I affected by CVE-2026-116467?** Check whether you run Ggml-Org Llama.Cpp at an affected version. Junglewise does that check for you around the clock: connect your code, cloud and domains, and it matches new vulnerabilities like this one against what you actually run and tells you when one reaches you.

## Technical details

The Mirostat Sampler component in src/llama-sampler.cpp fails to guard against non-finite (NaN or infinity) logits and probabilities, resulting in undefined behavior (float-cast-overflow) or assertion failures. Both Mirostat v1 and v2 variants are reachable remotely via the llama-server interface when mirostat=1 or mirostat=2, allowing an attacker to cause an abort or crash without authentication.

## Affected products

- ggml-org llama.cpp up to 0.7.0

## Timeline

- 2026-10-26: disclosed
- 2026-10-05: other: Pull request #29992 submitted with fix but not yet merged

## References

- [https://github.com/ggml-org/llama.cpp/](https://github.com/ggml-org/llama.cpp/)
- [https://github.com/ggml-org/llama.cpp/pull/29992](https://github.com/ggml-org/llama.cpp/pull/29992)

## Related threats

- [llama.cpp ggml-rpc server denial of service](https://junglewise.ai/threats/cve-2026-119052-llama-cpp-b2877-before-b11935-contains-an-uncaught-exception)
- [llama.cpp out-of-bounds write in ggml-rpc RDMA transport](https://junglewise.ai/threats/cve-2026-118375-llama-cpp-b8841-before-b11865-contains-an-out-of-bounds-write)
- [llama.cpp integer overflow in gguf_init_from_reader](https://junglewise.ai/threats/cve-2026-116829-llama-cpp-b10398-before-b11301-contains-an-integer-overflow)
- [llama.cpp heap out-of-bounds write in PLAMO2 tokenizer](https://junglewise.ai/threats/cve-2026-116816-llama-cpp-b5899-before-b10261-contains-a-heap-out-of-bounds)
- [llama.cpp heap out-of-bounds read in tokenizer loading](https://junglewise.ai/threats/cve-2025-71439-llama-cpp-before-b5649-contains-a-heap-out-of-bounds-read)